* src/vm/jit/trace.c (trace_java_call_enter): Removed unused variable.
[cacao.git] / src / vm / jit / replace.c
index 1eb166d0f216b523f28bad5de2fb3a92867f02d5..c0b170231bfd8be55c1e4c2b871644bebbf7fc5f 100644 (file)
@@ -1,6 +1,6 @@
-/* vm/jit/replace.c - on-stack replacement of methods
+/* src/vm/jit/replace.c - on-stack replacement of methods
 
-   Copyright (C) 1996-2005, 2006 R. Grafl, A. Krall, C. Kruegel,
+   Copyright (C) 1996-2005, 2006, 2007 R. Grafl, A. Krall, C. Kruegel,
    C. Oates, R. Obermaisser, M. Platter, M. Probst, S. Ring,
    E. Steiner, C. Thalinger, D. Thuernbeck, P. Tomsich, C. Ullrich,
    J. Wenninger, Institut f. Computersprachen - TU Wien
    Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
    02110-1301, USA.
 
-   Contact: cacao@cacaojvm.org
-
-   Authors: Edwin Steiner
-
-   Changes:
-
-   $Id$
-
 */
 
 #include "config.h"
 #include "vm/types.h"
 
 #include <assert.h>
+#include <stdint.h>
 #include <stdlib.h>
 
 #include "arch.h"
 
+#if defined(ENABLE_GC_CACAO)
+# include "mm/cacao-gc/gc.h"
+#endif
+
 #include "mm/memory.h"
+
+#include "threads/threads-common.h"
+
 #include "toolbox/logging.h"
-#include "vm/options.h"
+
 #include "vm/stringlocal.h"
+
 #include "vm/jit/abi.h"
-#include "vm/jit/jit.h"
-#include "vm/jit/replace.h"
-#include "vm/jit/stack.h"
 #include "vm/jit/asmpart.h"
 #include "vm/jit/disass.h"
-#include "vm/jit/show.h"
+#include "vm/jit/jit.h"
+#include "vm/jit/md.h"
 #include "vm/jit/methodheader.h"
+#include "vm/jit/replace.h"
+#include "vm/jit/show.h"
+#include "vm/jit/stack.h"
+
+#include "vmcore/options.h"
+#include "vmcore/classcache.h"
 
-#include "native/include/java_lang_String.h"
 
 #define REPLACE_PATCH_DYNAMIC_CALL
 /*#define REPLACE_PATCH_ALL*/
 
+#if defined(ENABLE_VMLOG)
+#include <vmlog_cacao.h>
+#endif
+
+/*** architecture-dependent configuration *************************************/
+
+/* first unset the macros (default) */
+#undef REPLACE_RA_BETWEEN_FRAMES
+#undef REPLACE_RA_TOP_OF_FRAME
+#undef REPLACE_RA_LINKAGE_AREA
+#undef REPLACE_LEAFMETHODS_RA_REGISTER
+#undef REPLACE_REG_RA
+
+/* i386, x86_64 and m68k */
+#if defined(__I386__) || defined(__X86_64__) || defined(__M68K__)
+#define REPLACE_RA_BETWEEN_FRAMES
+/* alpha */
+#elif defined(__ALPHA__)
+#define REPLACE_RA_TOP_OF_FRAME
+#define REPLACE_LEAFMETHODS_RA_REGISTER
+#define REPLACE_REG_RA REG_RA
+/* powerpc */
+#elif defined(__POWERPC__)
+#define REPLACE_RA_LINKAGE_AREA
+#define REPLACE_LEAFMETHODS_RA_REGISTER
+#define REPLACE_REG_RA REG_ITMP3 /* the execution state has the LR in itmp3 */
+/* s390 */
+#elif defined(__S390__)
+#define REPLACE_RA_TOP_OF_FRAME
+#define REPLACE_REG_RA REG_ITMP3
+#endif
+
 
 /*** configuration of native stack slot size **********************************/
 
@@ -76,25 +112,14 @@ typedef u8 stackslot_t;
 
 /*** debugging ****************************************************************/
 
-/*#define REPLACE_VERBOSE*/
-
 #if !defined(NDEBUG)
-void java_value_print(s4 type, u8 value);
-
-static char *replace_type_str[] = {
-       "STD",
-       "EXH",
-       "SBR",
-       "CALL",
-       "INLINE",
-       "RETURN",
-       "BODY"
-};
-#endif /* !defined(NDEBUG) */
+static void java_value_print(s4 type, replace_val_t value);
+static void replace_stackframeinfo_println(stackframeinfo_t *sfi);
+#endif
 
-#if !defined(NDEBUG) && defined(REPLACE_VERBOSE)
-#define DOLOG(code) do{ if (1) { code; } } while(0)
-#define DOLOG_SHORT(code) do{ if (1) { code; } } while(0)
+#if !defined(NDEBUG)
+#define DOLOG(code)        do{ if (opt_TraceReplacement > 1) { code; } } while(0)
+#define DOLOG_SHORT(code)  do{ if (opt_TraceReplacement > 0) { code; } } while(0)
 #else
 #define DOLOG(code)
 #define DOLOG_SHORT(code)
@@ -139,55 +164,7 @@ static int stat_dist_method_rplpoints[20] = { 0 };
         else (array)[limit]++;                                       \
     } while (0)
 
-#define REPLACE_PRINT_DIST(name, array) \
-       printf("    " name " distribution:\n"); \
-       print_freq(stdout, (array), sizeof(array)/sizeof(int) - 1);
-
-static void print_freq(FILE *file,int *array,int limit)
-{
-       int i;
-       int sum = 0;
-       int cum = 0;
-       for (i=0; i<limit; ++i)
-               sum += array[i];
-       sum += array[limit];
-       for (i=0; i<limit; ++i) {
-               cum += array[i];
-               fprintf(file,"      %3d: %8d (cum %3d%%)\n",
-                               i, array[i], (sum) ? ((100*cum)/sum) : 0);
-       }
-       fprintf(file,"    >=%3d: %8d\n",limit,array[limit]);
-}
-
-void replace_print_statistics(void)
-{
-       printf("replacement statistics:\n");
-       printf("    # of replacements:   %d\n", stat_replacements);
-       printf("    # of frames:         %d\n", stat_frames);
-       printf("    # of recompilations: %d\n", stat_recompile);
-       printf("    patched static calls:%d\n", stat_staticpatch);
-       printf("    unrolled inlines:    %d\n", stat_unroll_inline);
-       printf("    unrolled calls:      %d\n", stat_unroll_call);
-       REPLACE_PRINT_DIST("frame depth", stat_dist_frames);
-       REPLACE_PRINT_DIST("locals per frame", stat_dist_locals);
-       REPLACE_PRINT_DIST("ADR locals per frame", stat_dist_locals_adr);
-       REPLACE_PRINT_DIST("primitive locals per frame", stat_dist_locals_prim);
-       REPLACE_PRINT_DIST("RET locals per frame", stat_dist_locals_ret);
-       REPLACE_PRINT_DIST("void locals per frame", stat_dist_locals_void);
-       REPLACE_PRINT_DIST("stack slots per frame", stat_dist_stack);
-       REPLACE_PRINT_DIST("ADR stack slots per frame", stat_dist_stack_adr);
-       REPLACE_PRINT_DIST("primitive stack slots per frame", stat_dist_stack_prim);
-       REPLACE_PRINT_DIST("RET stack slots per frame", stat_dist_stack_ret);
-       printf("\n");
-       printf("    # of methods:            %d\n", stat_methods);
-       printf("    # of replacement points: %d\n", stat_rploints);
-       printf("    # of regallocs:          %d\n", stat_regallocs);
-       printf("        per rplpoint:        %f\n", (double)stat_regallocs / stat_rploints);
-       printf("        per method:          %f\n", (double)stat_regallocs / stat_methods);
-       REPLACE_PRINT_DIST("replacement points per method", stat_dist_method_rplpoints);
-       printf("\n");
-
-}
+static void replace_statistics_source_frame(sourceframe_t *frame);
 
 #else
 
@@ -198,6 +175,7 @@ void replace_print_statistics(void)
 
 #endif /* defined(REPLACE_STATISTICS) */
 
+
 /*** constants used internally ************************************************/
 
 #define TOP_IS_NORMAL    0
@@ -206,6 +184,11 @@ void replace_print_statistics(void)
 #define TOP_IS_VOID      3
 
 
+/******************************************************************************/
+/* PART I: Creating / freeing replacement points                              */
+/******************************************************************************/
+
+
 /* replace_create_replacement_point ********************************************
  
    Create a replacement point.
@@ -249,16 +232,13 @@ static void replace_create_replacement_point(jitdata *jd,
 
        rp->method = (iinfo) ? iinfo->method : jd->m;
        rp->pc = NULL;        /* set by codegen */
-       rp->outcode = NULL;   /* set by codegen */
        rp->callsize = 0;     /* set by codegen */
-       rp->target = NULL;
        rp->regalloc = ra;
        rp->flags = 0;
        rp->type = type;
        rp->id = iptr->flags.bits >> INS_FLAG_ID_SHIFT;
 
        /* XXX unify these two fields */
-       rp->code = jd->code;
        rp->parent = (iinfo) ? iinfo->rp : NULL;
 
        /* store local allocation info of javalocals */
@@ -270,17 +250,17 @@ static void replace_create_replacement_point(jitdata *jd,
                                continue;
 
                        ra->index = i;
-                       if (index < UNUSED) {
-                               ra->regoff = (UNUSED - index) - 1;
-                               ra->type = TYPE_RET;
-                               ra->flags = 0;
-                       }
-                       else {
+                       if (index >= 0) {
                                v = VAR(index);
                                ra->flags = v->flags & (INMEMORY);
                                ra->regoff = v->vv.regoff;
                                ra->type = v->type;
                        }
+                       else {
+                               ra->regoff = RETADDR_FROM_JAVALOCAL(index);
+                               ra->type = TYPE_RET;
+                               ra->flags = 0;
+                       }
                        ra++;
                }
        }
@@ -493,6 +473,17 @@ bool replace_create_replacement_points(jitdata *jd)
 
                for (; iptr != iend; ++iptr) {
                        switch (iptr->opc) {
+#if defined(ENABLE_GC_CACAO)
+                               case ICMD_BUILTIN:
+                                       md = iptr->sx.s23.s3.bte->md;
+                                       count++;
+                                       COUNT_javalocals(javalocals, m, alloccount);
+                                       alloccount += iptr->s1.argcount;
+                                       if (iinfo)
+                                               alloccount -= iinfo->throughcount;
+                                       break;
+#endif
+
                                case ICMD_INVOKESTATIC:
                                case ICMD_INVOKESPECIAL:
                                case ICMD_INVOKEVIRTUAL:
@@ -632,6 +623,9 @@ bool replace_create_replacement_points(jitdata *jd)
                        replace_create_replacement_point(jd, iinfo, rp++,
                                        bptr->type, bptr->iinstr, &ra,
                                        bptr->javalocals, bptr->invars + i, bptr->indepth - i, 0);
+
+                       if (JITDATA_HAS_FLAG_COUNTDOWN(jd))
+                               rp[-1].flags |= RPLPOINT_FLAG_COUNTDOWN;
                }
 
                /* iterate over the instructions */
@@ -641,6 +635,19 @@ bool replace_create_replacement_points(jitdata *jd)
 
                for (; iptr != iend; ++iptr) {
                        switch (iptr->opc) {
+#if defined(ENABLE_GC_CACAO)
+                               case ICMD_BUILTIN:
+                                       md = iptr->sx.s23.s3.bte->md;
+
+                                       i = (iinfo) ? iinfo->throughcount : 0;
+                                       replace_create_replacement_point(jd, iinfo, rp++,
+                                                       RPLPOINT_TYPE_CALL, iptr, &ra,
+                                                       javalocals, iptr->sx.s23.s2.args,
+                                                       iptr->s1.argcount - i,
+                                                       md->paramcount);
+                                       break;
+#endif
+
                                case ICMD_INVOKESTATIC:
                                case ICMD_INVOKESPECIAL:
                                case ICMD_INVOKEVIRTUAL:
@@ -728,6 +735,9 @@ bool replace_create_replacement_points(jitdata *jd)
        code->globalcount   = 0;
        code->savedintcount = INT_SAV_CNT - rd->savintreguse;
        code->savedfltcount = FLT_SAV_CNT - rd->savfltreguse;
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+       code->savedadrcount = ADR_SAV_CNT - rd->savadrreguse;
+#endif
        code->memuse        = rd->memuse;
        code->stackframesize = jd->cd->stackframesize;
 
@@ -767,66 +777,186 @@ void replace_free_replacement_points(codeinfo *code)
 }
 
 
-/* replace_activate_replacement_point ******************************************
+/******************************************************************************/
+/* PART II: Activating / deactivating replacement points                      */
+/******************************************************************************/
+
+
+/* replace_activate_replacement_points *****************************************
  
-   Activate a replacement point. When this function returns, the
-   replacement point is "armed", that is each thread reaching this point
-   will be replace to `target`.
+   Activate the replacement points of the given compilation unit. When this
+   function returns, the replacement points are "armed", so each thread
+   reaching one of the points will enter the replacement mechanism.
    
    IN:
-       rp...............replacement point to activate
-          target...........target of replacement
+       code.............codeinfo of which replacement points should be
+                                               activated
+          mappable.........if true, only mappable replacement points are
+                                               activated
   
 *******************************************************************************/
 
-void replace_activate_replacement_point(rplpoint *rp,rplpoint *target)
+void replace_activate_replacement_points(codeinfo *code, bool mappable)
 {
-       assert(rp->target == NULL);
+       rplpoint *rp;
+       s4        i;
+       s4        count;
+       u1       *savedmcode;
+
+       assert(code->savedmcode == NULL);
+
+       /* count trappable replacement points */
+
+       count = 0;
+       i = code->rplpointcount;
+       rp = code->rplpoints;
+       for (; i--; rp++) {
+               if (rp->flags & RPLPOINT_FLAG_NOTRAP)
+                       continue;
+
+               if (mappable && (rp->type == RPLPOINT_TYPE_RETURN))
+                       continue;
+
+               count++;
+       }
+
+       /* allocate buffer for saved machine code */
+
+       savedmcode = MNEW(u1, count * REPLACEMENT_PATCH_SIZE);
+       code->savedmcode = savedmcode;
+       savedmcode += count * REPLACEMENT_PATCH_SIZE;
+
+       /* activate trappable replacement points */
+       /* (in reverse order to handle overlapping points within basic blocks) */
+
+       i = code->rplpointcount;
+       rp = code->rplpoints + i;
+       while (rp--, i--) {
+               assert(!(rp->flags & RPLPOINT_FLAG_ACTIVE));
+
+               if (rp->flags & RPLPOINT_FLAG_NOTRAP)
+                       continue;
 
-       DOLOG( printf("activate replacement point:\n");
-                  replace_replacement_point_println(rp, 1); fflush(stdout); );
+               if (mappable && (rp->type == RPLPOINT_TYPE_RETURN))
+                       continue;
+
+               DOLOG( printf("activate replacement point:\n");
+                          replace_replacement_point_println(rp, 1); fflush(stdout); );
+
+               savedmcode -= REPLACEMENT_PATCH_SIZE;
+
+#if defined(ENABLE_JIT)
+# if defined(ENABLE_DISASSEMBLER)
+               DOLOG( printf("\tinstruction before: ");
+                          disassinstr(rp->pc); fflush(stdout); );
+# endif
 
-       rp->target = target;
+               md_patch_replacement_point(rp->pc, savedmcode, false);
 
-#if (defined(__I386__) || defined(__X86_64__) || defined(__ALPHA__) || defined(__POWERPC__) || defined(__MIPS__)) && defined(ENABLE_JIT)
-       md_patch_replacement_point(rp);
+# if defined(ENABLE_DISASSEMBLER)
+               DOLOG( printf("\tinstruction after : ");
+                          disassinstr(rp->pc); fflush(stdout); );
+# endif
 #endif
+
+               rp->flags |= RPLPOINT_FLAG_ACTIVE;
+       }
+
+       assert(savedmcode == code->savedmcode);
 }
 
 
-/* replace_deactivate_replacement_point ****************************************
+/* replace_deactivate_replacement_point***************************************
  
-   Deactivate a replacement point. When this function returns, the
-   replacement point is "un-armed", that is a each thread reaching this point
-   will just continue normally.
+   Deactivate a replacement points in the given compilation unit.
+   When this function returns, the replacement points will be "un-armed",
+   that is a each thread reaching a point will just continue normally.
    
    IN:
-       rp...............replacement point to deactivate
+       code.............the compilation unit
   
 *******************************************************************************/
 
-void replace_deactivate_replacement_point(rplpoint *rp)
+void replace_deactivate_replacement_points(codeinfo *code)
 {
-       assert(rp->target);
+       rplpoint *rp;
+       s4        i;
+       s4        count;
+       u1       *savedmcode;
+
+       if (code->savedmcode == NULL) {
+               /* disarm countdown points by patching the branches */
+
+               i = code->rplpointcount;
+               rp = code->rplpoints;
+               for (; i--; rp++) {
+                       if ((rp->flags & (RPLPOINT_FLAG_ACTIVE | RPLPOINT_FLAG_COUNTDOWN))
+                                       == RPLPOINT_FLAG_COUNTDOWN)
+                       {
+#if 0
+                               *(s4*) (rp->pc + 9) = 0; /* XXX machine dependent! */
+#endif
+                       }
+               }
+               return;
+       }
+
+       assert(code->savedmcode != NULL);
+       savedmcode = code->savedmcode;
+
+       /* de-activate each trappable replacement point */
 
-       DOLOG( printf("deactivate replacement point:\n");
-                  replace_replacement_point_println(rp, 1); fflush(stdout); );
+       i = code->rplpointcount;
+       rp = code->rplpoints;
+       count = 0;
+       for (; i--; rp++) {
+               if (!(rp->flags & RPLPOINT_FLAG_ACTIVE))
+                       continue;
 
-       rp->target = NULL;
+               count++;
 
-#if (defined(__I386__) || defined(__X86_64__) || defined(__ALPHA__) || defined(__POWERPC__) || defined(__MIPS__)) && defined(ENABLE_JIT)
-       md_patch_replacement_point(rp);
+               DOLOG( printf("deactivate replacement point:\n");
+                          replace_replacement_point_println(rp, 1); fflush(stdout); );
+
+#if defined(ENABLE_JIT)
+# if defined(ENABLE_DISASSEMBLER)
+               DOLOG( printf("\tinstruction before: ");
+                          disassinstr(rp->pc); fflush(stdout); );
+# endif
+
+               md_patch_replacement_point(rp->pc, savedmcode, true);
+
+# if defined(ENABLE_DISASSEMBLER)
+               DOLOG( printf("\tinstruction before: ");
+                          disassinstr(rp->pc); fflush(stdout); );
+# endif
 #endif
+
+               rp->flags &= ~RPLPOINT_FLAG_ACTIVE;
+
+               savedmcode += REPLACEMENT_PATCH_SIZE;
+       }
+
+       assert(savedmcode == code->savedmcode + count * REPLACEMENT_PATCH_SIZE);
+
+       /* free saved machine code */
+
+       MFREE(code->savedmcode, u1, count * REPLACEMENT_PATCH_SIZE);
+       code->savedmcode = NULL;
 }
 
 
+/******************************************************************************/
+/* PART III: The replacement mechanism                                        */
+/******************************************************************************/
+
+
 /* replace_read_value **********************************************************
 
    Read a value with the given allocation from the execution state.
    
    IN:
           es...............execution state
-          sp...............stack pointer of the execution state (XXX eliminate?)
           ra...............allocation
           javaval..........where to put the value
 
@@ -836,19 +966,18 @@ void replace_deactivate_replacement_point(rplpoint *rp)
 *******************************************************************************/
 
 static void replace_read_value(executionstate_t *es,
-                                                          stackslot_t *sp,
                                                           rplalloc *ra,
-                                                          u8 *javaval)
+                                                          replace_val_t *javaval)
 {
        if (ra->flags & INMEMORY) {
                /* XXX HAS_4BYTE_STACKSLOT may not be the right discriminant here */
 #ifdef HAS_4BYTE_STACKSLOT
                if (IS_2_WORD_TYPE(ra->type)) {
-                       *javaval = *(u8*)(sp + ra->regoff);
+                       javaval->l = *(u8*)(es->sp + ra->regoff);
                }
                else {
 #endif
-                       *javaval = sp[ra->regoff];
+                       javaval->p = *(ptrint*)(es->sp + ra->regoff);
 #ifdef HAS_4BYTE_STACKSLOT
                }
 #endif
@@ -856,10 +985,25 @@ static void replace_read_value(executionstate_t *es,
        else {
                /* allocated register */
                if (IS_FLT_DBL_TYPE(ra->type)) {
-                       *javaval = es->fltregs[ra->regoff];
+                       javaval->d = es->fltregs[ra->regoff];
+
+                       if (ra->type == TYPE_FLT)
+                               javaval->f = javaval->d;
                }
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+               else if (IS_ADR_TYPE(ra->type)) {
+                       javaval->p = es->adrregs[ra->regoff];
+               }
+#endif
                else {
-                       *javaval = es->intregs[ra->regoff];
+#if defined(SUPPORT_COMBINE_INTEGER_REGISTERS)
+                       if (ra->type == TYPE_LNG) {
+                               javaval->words.lo = es->intregs[GET_LOW_REG(ra->regoff)];
+                               javaval->words.hi = es->intregs[GET_HIGH_REG(ra->regoff)];
+                       }
+                       else
+#endif /* defined(SUPPORT_COMBINE_INTEGER_REGISTERS) */
+                               javaval->p = es->intregs[ra->regoff];
                }
        }
 }
@@ -871,53 +1015,99 @@ static void replace_read_value(executionstate_t *es,
    
    IN:
           es...............execution state
-          sp...............stack pointer of the execution state (XXX eliminate?)
           ra...............allocation
           *javaval.........the value
 
 *******************************************************************************/
 
 static void replace_write_value(executionstate_t *es,
-                                                           stackslot_t *sp,
                                                            rplalloc *ra,
-                                                           u8 *javaval)
+                                                           replace_val_t *javaval)
 {
        if (ra->flags & INMEMORY) {
                /* XXX HAS_4BYTE_STACKSLOT may not be the right discriminant here */
 #ifdef HAS_4BYTE_STACKSLOT
                if (IS_2_WORD_TYPE(ra->type)) {
-                       *(u8*)(sp + ra->regoff) = *javaval;
+                       *(u8*)(es->sp + ra->regoff) = javaval->l;
                }
                else {
 #endif
-                       sp[ra->regoff] = *javaval;
+                       *(ptrint*)(es->sp + ra->regoff) = javaval->p;
 #ifdef HAS_4BYTE_STACKSLOT
                }
 #endif
        }
        else {
                /* allocated register */
-               if (IS_FLT_DBL_TYPE(ra->type)) {
-                       es->fltregs[ra->regoff] = *javaval;
-               }
-               else {
-                       es->intregs[ra->regoff] = *javaval;
+               switch (ra->type) {
+                       case TYPE_FLT:
+                               es->fltregs[ra->regoff] = (double) javaval->f;
+                               break;
+                       case TYPE_DBL:
+                               es->fltregs[ra->regoff] = javaval->d;
+                               break;
+#if defined(SUPPORT_COMBINE_INTEGER_REGISTERS)
+                       case TYPE_LNG:
+                               es->intregs[GET_LOW_REG(ra->regoff)] = javaval->words.lo;
+                               es->intregs[GET_HIGH_REG(ra->regoff)] = javaval->words.hi;
+                               break;
+#endif
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+                       case TYPE_ADR:
+                               es->adrregs[ra->regoff] = javaval->p;
+#endif
+                       default:
+                               es->intregs[ra->regoff] = javaval->p;
                }
        }
 }
 
 
-/* replace_read_executionstate *************************************************
+/* replace_new_sourceframe *****************************************************
 
-   Read the given executions state and translate it to a source frame.
+   Allocate a new source frame and insert it at the front of the frame list.
    
+   IN:
+          ss...............the source state
+
+   OUT:
+          ss->frames.......set to new frame (the new head of the frame list).
+
+   RETURN VALUE:
+       returns the new frame
+
+*******************************************************************************/
+
+static sourceframe_t *replace_new_sourceframe(sourcestate_t *ss)
+{
+       sourceframe_t *frame;
+
+       frame = DNEW(sourceframe_t);
+       MZERO(frame, sourceframe_t, 1);
+
+       frame->down = ss->frames;
+       ss->frames = frame;
+
+       return frame;
+}
+
+
+/* replace_read_executionstate *************************************************
+
+   Read a source frame from the given executions state.
+   The new source frame is pushed to the front of the frame list of the
+   source state.
+
    IN:
        rp...............replacement point at which `es` was taken
           es...............execution state
-          ss...............where to put the source state
+          ss...............the source state to add the source frame to
+          topframe.........true, if the first (top-most) source frame on the
+                           stack is to be read
 
    OUT:
-       *ss..............the source state derived from the execution state
+       *ss..............the source state with the newly created source frame
+                           added
   
 *******************************************************************************/
 
@@ -931,9 +1121,10 @@ static s4 replace_normalize_type_map[] = {
 /* RPLPOINT_TYPE_BODY   |--> */ RPLPOINT_TYPE_STD
 };
 
+
 static void replace_read_executionstate(rplpoint *rp,
                                                                                executionstate_t *es,
-                                                                               sourcestate_t *ss,
+                                                                               sourcestate_t *ss,
                                                                                bool topframe)
 {
        methodinfo    *m;
@@ -946,7 +1137,7 @@ static void replace_read_executionstate(rplpoint *rp,
        stackslot_t   *sp;
        stackslot_t   *basesp;
 
-       code = rp->code;
+       code = code_find_codeinfo_for_pc(rp->pc);
        m = rp->method;
        topslot = TOP_IS_NORMAL;
 
@@ -962,42 +1153,39 @@ static void replace_read_executionstate(rplpoint *rp,
 
        /* calculate base stack pointer */
 
-       basesp = sp + code_get_stack_frame_size(code);
+       basesp = sp + code->stackframesize;
 
        /* create the source frame */
 
-       frame = DNEW(sourceframe_t);
-       frame->down = ss->frames;
+       frame = replace_new_sourceframe(ss);
        frame->method = rp->method;
        frame->id = rp->id;
        assert(rp->type >= 0 && rp->type < sizeof(replace_normalize_type_map)/sizeof(s4));
        frame->type = replace_normalize_type_map[rp->type];
-       frame->instance = 0;
-       frame->syncslotcount = 0;
-       frame->syncslots = NULL;
-       frame->readrp = rp;
-
-       ss->frames = frame;
+       frame->fromrp = rp;
+       frame->fromcode = code;
 
        /* read local variables */
 
        count = m->maxlocals;
        frame->javalocalcount = count;
-       frame->javalocals = DMNEW(u8, count);
+       frame->javalocals = DMNEW(replace_val_t, count);
        frame->javalocaltype = DMNEW(u1, count);
 
-#if !defined(NDEBUG)
        /* mark values as undefined */
        for (i=0; i<count; ++i) {
-               frame->javalocals[i] = (u8) 0x00dead0000dead00ULL;
+#if !defined(NDEBUG)
+               frame->javalocals[i].l = (u8) 0x00dead0000dead00ULL;
+#endif
                frame->javalocaltype[i] = TYPE_VOID;
        }
 
        /* some entries in the intregs array are not meaningful */
        /*es->intregs[REG_ITMP3] = (u8) 0x11dead1111dead11ULL;*/
-       es->intregs[REG_SP   ] = (u8) 0x11dead1111dead11ULL;
+#if !defined(NDEBUG)
+       es->intregs[REG_SP   ] = (ptrint) 0x11dead1111dead11ULL;
 #ifdef REG_PV
-       es->intregs[REG_PV   ] = (u8) 0x11dead1111dead11ULL;
+       es->intregs[REG_PV   ] = (ptrint) 0x11dead1111dead11ULL;
 #endif
 #endif /* !defined(NDEBUG) */
 
@@ -1010,9 +1198,9 @@ static void replace_read_executionstate(rplpoint *rp,
                assert(i < m->maxlocals);
                frame->javalocaltype[i] = ra->type;
                if (ra->type == TYPE_RET)
-                       frame->javalocals[i] = ra->regoff;
+                       frame->javalocals[i].i = ra->regoff;
                else
-                       replace_read_value(es, sp, ra, frame->javalocals + i);
+                       replace_read_value(es, ra, frame->javalocals + i);
                ra++;
                count--;
        }
@@ -1021,6 +1209,12 @@ static void replace_read_executionstate(rplpoint *rp,
 
 #if defined(REPLACE_PATCH_DYNAMIC_CALL)
        if (topframe && !(rp->method->flags & ACC_STATIC) && rp == code->rplpoints) {
+#if 1
+               /* we are at the start of the method body, so if local 0 is set, */
+               /* it is the instance.                                           */
+               if (frame->javalocaltype[0] == TYPE_ADR)
+                       frame->instance = frame->javalocals[0];
+#else
                rplalloc instra;
                methoddesc *md;
 
@@ -1031,25 +1225,26 @@ static void replace_read_executionstate(rplpoint *rp,
                instra.regoff = md->params[0].regoff;
                if (md->params[0].inmemory) {
                        instra.flags = INMEMORY;
-                       instra.regoff += (1 + code->stackframesize);
+                       instra.regoff += (1 + code->stackframesize) * SIZE_OF_STACKSLOT;
                }
                else {
                        instra.flags = 0;
                }
-               replace_read_value(es, sp, &instra, &(frame->instance));
+               replace_read_value(es, &instra, &(frame->instance));
+#endif
        }
 #endif /* defined(REPLACE_PATCH_DYNAMIC_CALL) */
 
        /* read stack slots */
 
        frame->javastackdepth = count;
-       frame->javastack = DMNEW(u8, count);
+       frame->javastack = DMNEW(replace_val_t, count);
        frame->javastacktype = DMNEW(u1, count);
 
 #if !defined(NDEBUG)
        /* mark values as undefined */
        for (i=0; i<count; ++i) {
-               frame->javastack[i] = (u8) 0x00dead0000dead00ULL;
+               frame->javastack[i].l = (u8) 0x00dead0000dead00ULL;
                frame->javastacktype[i] = TYPE_VOID;
        }
 #endif /* !defined(NDEBUG) */
@@ -1062,7 +1257,8 @@ static void replace_read_executionstate(rplpoint *rp,
                assert(count);
 
                assert(ra->index == RPLALLOC_STACK);
-               frame->javastack[i] = sp[-1];
+               assert(ra->type == TYPE_ADR);
+               frame->javastack[i].p = sp[-1];
                frame->javastacktype[i] = TYPE_ADR; /* XXX RET */
                count--;
                i++;
@@ -1072,7 +1268,8 @@ static void replace_read_executionstate(rplpoint *rp,
                assert(count);
 
                assert(ra->index == RPLALLOC_STACK);
-               frame->javastack[i] = es->intregs[REG_ITMP1];
+               assert(ra->type == TYPE_ADR);
+               frame->javastack[i].p = es->intregs[REG_ITMP1];
                frame->javastacktype[i] = TYPE_ADR; /* XXX RET */
                count--;
                i++;
@@ -1082,7 +1279,7 @@ static void replace_read_executionstate(rplpoint *rp,
                assert(count);
 
                assert(ra->index == RPLALLOC_STACK);
-               frame->javastack[i] = 0;
+               frame->javastack[i].l = 0;
                frame->javastacktype[i] = TYPE_VOID;
                count--;
                i++;
@@ -1104,8 +1301,8 @@ static void replace_read_executionstate(rplpoint *rp,
                                assert(calleeframe->syncslots == NULL);
 
                                calleeframe->syncslotcount = 1;
-                               calleeframe->syncslots = DMNEW(u8, 1);
-                               replace_read_value(es,sp,ra,calleeframe->syncslots);
+                               calleeframe->syncslots = DMNEW(replace_val_t, 1);
+                               replace_read_value(es,ra,calleeframe->syncslots);
                        }
 
                        frame->javastackdepth--;
@@ -1121,9 +1318,9 @@ static void replace_read_executionstate(rplpoint *rp,
                }
                else {
                        if (ra->type == TYPE_RET)
-                               frame->javastack[i] = ra->regoff;
+                               frame->javastack[i].i = ra->regoff;
                        else
-                               replace_read_value(es,sp,ra,frame->javastack + i);
+                               replace_read_value(es,ra,frame->javastack + i);
                        frame->javastacktype[i] = ra->type;
                        i++;
                }
@@ -1133,13 +1330,16 @@ static void replace_read_executionstate(rplpoint *rp,
 
 /* replace_write_executionstate ************************************************
 
-   Translate the given source state into an execution state.
-   
+   Pop a source frame from the front of the frame list of the given source state
+   and write its values into the execution state.
+
    IN:
        rp...............replacement point for which execution state should be
-                           creates
-          es...............where to put the execution state
+                           created
+          es...............the execution state to modify
           ss...............the given source state
+          topframe.........true, if this is the last (top-most) source frame to be
+                           translated
 
    OUT:
        *es..............the execution state derived from the source state
@@ -1161,7 +1361,7 @@ static void replace_write_executionstate(rplpoint *rp,
        stackslot_t    *sp;
        stackslot_t    *basesp;
 
-       code = rp->code;
+       code = code_find_codeinfo_for_pc(rp->pc);
        m = rp->method;
        topslot = TOP_IS_NORMAL;
 
@@ -1175,7 +1375,7 @@ static void replace_write_executionstate(rplpoint *rp,
 
        sp = (stackslot_t *) es->sp;
 
-       basesp = sp + code_get_stack_frame_size(code);
+       basesp = sp + code->stackframesize;
 
        /* in some cases the top stack slot is passed in REG_ITMP1 */
 
@@ -1196,7 +1396,7 @@ static void replace_write_executionstate(rplpoint *rp,
                        /* XXX assert that it matches this rplpoint */
                }
                else
-                       replace_write_value(es, sp, ra, frame->javalocals + i);
+                       replace_write_value(es, ra, frame->javalocals + i);
                count--;
                ra++;
        }
@@ -1213,7 +1413,7 @@ static void replace_write_executionstate(rplpoint *rp,
                assert(ra->index == RPLALLOC_STACK);
                assert(i < frame->javastackdepth);
                assert(frame->javastacktype[i] == TYPE_ADR);
-               sp[-1] = frame->javastack[i];
+               sp[-1] = frame->javastack[i].p;
                count--;
                i++;
                ra++;
@@ -1224,7 +1424,7 @@ static void replace_write_executionstate(rplpoint *rp,
                assert(ra->index == RPLALLOC_STACK);
                assert(i < frame->javastackdepth);
                assert(frame->javastacktype[i] == TYPE_ADR);
-               es->intregs[REG_ITMP1] = frame->javastack[i];
+               es->intregs[REG_ITMP1] = frame->javastack[i].p;
                count--;
                i++;
                ra++;
@@ -1253,7 +1453,7 @@ static void replace_write_executionstate(rplpoint *rp,
                                assert(frame->down->syncslotcount == 1); /* XXX need to understand more cases */
                                assert(frame->down->syncslots != NULL);
 
-                               replace_write_value(es,sp,ra,frame->down->syncslots);
+                               replace_write_value(es,ra,frame->down->syncslots);
                        }
                        continue;
                }
@@ -1272,7 +1472,7 @@ static void replace_write_executionstate(rplpoint *rp,
                                /* XXX assert that it matches this rplpoint */
                        }
                        else {
-                               replace_write_value(es,sp,ra,frame->javastack + i);
+                               replace_write_value(es,ra,frame->javastack + i);
                        }
                        i++;
                }
@@ -1317,26 +1517,17 @@ u1* replace_pop_activation_record(executionstate_t *es,
 
        /* read the return address */
 
-       ra = md_stacktrace_get_returnaddress(es->sp,
-                       SIZE_OF_STACKSLOT * es->code->stackframesize);
-
-       DOLOG( printf("return address: %p\n", (void*)ra); );
-
-       /* find the new codeinfo */
-
-       pv = md_codegen_get_pv_from_pc(ra);
-
-       DOLOG( printf("PV = %p\n", (void*) pv); );
-
-       if (pv == NULL)
-               return NULL;
-
-       code = *(codeinfo **)(pv + CodeinfoPointer);
+#if defined(REPLACE_LEAFMETHODS_RA_REGISTER)
+       if (code_is_leafmethod(es->code))
+               ra = (u1*) (ptrint) es->intregs[REPLACE_REG_RA];
+       else
+#endif
+               ra = md_stacktrace_get_returnaddress(es->sp,
+                               SIZE_OF_STACKSLOT * es->code->stackframesize);
 
-       DOLOG( printf("CODE = %p\n", (void*) code); );
+       DOLOG( printf("RA = %p\n", (void*)ra); );
 
-       if (code == NULL)
-               return NULL;
+       assert(ra);
 
        /* calculate the base of the stack frame */
 
@@ -1349,11 +1540,27 @@ u1* replace_pop_activation_record(executionstate_t *es,
        assert(frame->syncslots == NULL);
        count = code_get_sync_slot_count(es->code);
        frame->syncslotcount = count;
-       frame->syncslots = DMNEW(u8, count);
+       frame->syncslots = DMNEW(replace_val_t, count);
        for (i=0; i<count; ++i) {
-               frame->syncslots[i] = sp[es->code->memuse + i];
+               frame->syncslots[i].p = sp[es->code->memuse + i]; /* XXX */
        }
 
+       /* restore return address, if part of frame */
+
+#if defined(REPLACE_RA_TOP_OF_FRAME)
+#if defined(REPLACE_LEAFMETHODS_RA_REGISTER)
+       if (!code_is_leafmethod(es->code))
+#endif
+               es->intregs[REPLACE_REG_RA] = *--basesp;
+#endif /* REPLACE_RA_TOP_OF_FRAME */
+
+#if defined(REPLACE_RA_LINKAGE_AREA)
+#if defined(REPLACE_LEAFMETHODS_RA_REGISTER)
+       if (!code_is_leafmethod(es->code))
+#endif
+               es->intregs[REPLACE_REG_RA] = basesp[LA_LR_OFFSET / sizeof(stackslot_t)];
+#endif /* REPLACE_RA_LINKAGE_AREA */
+
        /* restore saved int registers */
 
        reg = INT_REG_CNT;
@@ -1371,33 +1578,201 @@ u1* replace_pop_activation_record(executionstate_t *es,
                while (nregdescfloat[--reg] != REG_SAV)
                        ;
                basesp -= STACK_SLOTS_PER_FLOAT;
-               es->fltregs[reg] = *(u8*)basesp;
+               es->fltregs[reg] = *(double*)basesp;
        }
 
-       /* Set the new pc. Subtract one so we do not hit the replacement point */
-       /* of the instruction following the call, if there is one.             */
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+       /* restore saved adr registers */
 
-       es->pc = ra - 1;
+       reg = ADR_REG_CNT;
+       for (i=0; i<es->code->savedadrcount; ++i) {
+               while (nregdescadr[--reg] != REG_SAV)
+                       ;
+               es->adrregs[reg] = *--basesp;
+       }
+#endif
 
        /* adjust the stackpointer */
 
        es->sp += SIZE_OF_STACKSLOT * es->code->stackframesize;
+
+#if defined(REPLACE_RA_BETWEEN_FRAMES)
        es->sp += SIZE_OF_STACKSLOT; /* skip return address */
+#endif
+
+       /* Set the new pc. Subtract one so we do not hit the replacement point */
+       /* of the instruction following the call, if there is one.             */
+
+       es->pc = ra - 1;
+
+       /* find the new codeinfo */
+
+       pv = md_codegen_get_pv_from_pc(ra);
+       DOLOG( printf("PV = %p\n", (void*) pv); );
+
+       code = code_get_codeinfo_for_pv(pv);
+       DOLOG( printf("CODE = %p\n", (void*) code); );
+
+       /* return NULL if we reached native code */
 
        es->pv = pv;
        es->code = code;
 
+       /* in debugging mode clobber non-saved registers */
+
 #if !defined(NDEBUG)
        /* for debugging */
        for (i=0; i<INT_REG_CNT; ++i)
-               if (nregdescint[i] != REG_SAV)
-                       es->intregs[i] = 0x33dead3333dead33ULL;
+               if ((nregdescint[i] != REG_SAV)
+#if defined(REG_RA)
+                               && (i != REPLACE_REG_RA)
+#endif
+                       )
+                       es->intregs[i] = (ptrint) 0x33dead3333dead33ULL;
        for (i=0; i<FLT_REG_CNT; ++i)
                if (nregdescfloat[i] != REG_SAV)
-                       es->fltregs[i] = 0x33dead3333dead33ULL;
+                       *(u8*)&(es->fltregs[i]) = 0x33dead3333dead33ULL;
+# if defined(HAS_ADDRESS_REGISTER_FILE)
+       for (i=0; i<ADR_REG_CNT; ++i)
+               if (nregdescadr[i] != REG_SAV)
+                       es->adrregs[i] = (ptrint) 0x33dead3333dead33ULL;
+# endif
 #endif /* !defined(NDEBUG) */
 
-       return ra;
+       return (code) ? ra : NULL;
+}
+
+
+/* replace_patch_method_pointer ************************************************
+
+   Patch a method pointer (may be in code, data segment, vftbl, or interface
+   table).
+
+   IN:
+          mpp..............address of the method pointer to patch
+          entrypoint.......the new entrypoint of the method
+          kind.............kind of call to patch, used only for debugging
+
+*******************************************************************************/
+
+static void replace_patch_method_pointer(methodptr *mpp,
+                                                                                methodptr entrypoint,
+                                                                                const char *kind)
+{
+#if !defined(NDEBUG)
+       codeinfo       *oldcode;
+       codeinfo       *newcode;
+#endif
+
+       DOLOG( printf("patch method pointer from: %p to %p\n",
+                                 (void*) *mpp, (void*)entrypoint); );
+
+#if !defined(NDEBUG)
+       oldcode = code_get_codeinfo_for_pv(*mpp);
+       newcode = code_get_codeinfo_for_pv(entrypoint);
+
+       DOLOG_SHORT( printf("\tpatch %s %p ", kind, (void*) oldcode);
+                                method_println(oldcode->m);
+                                printf("\t      with      %p ", (void*) newcode);
+                                method_println(newcode->m); );
+
+       assert(oldcode->m == newcode->m);
+#endif
+
+       /* write the new entrypoint */
+
+       *mpp = (methodptr) entrypoint;
+}
+
+
+/* replace_patch_class *********************************************************
+
+   Patch a method in the given class.
+
+   IN:
+          vftbl............vftbl of the class
+          m................the method to patch
+          oldentrypoint....the old entrypoint to replace
+          entrypoint.......the new entrypoint
+
+*******************************************************************************/
+
+void replace_patch_class(vftbl_t *vftbl,
+                                                methodinfo *m,
+                                                u1 *oldentrypoint,
+                                                u1 *entrypoint)
+{
+       s4                 i;
+       methodptr         *mpp;
+       methodptr         *mppend;
+
+       /* patch the vftbl of the class */
+
+       replace_patch_method_pointer(vftbl->table + m->vftblindex,
+                                                                entrypoint,
+                                                                "virtual  ");
+
+       /* patch the interface tables */
+
+       assert(oldentrypoint);
+
+       for (i=0; i < vftbl->interfacetablelength; ++i) {
+               mpp = vftbl->interfacetable[-i];
+               mppend = mpp + vftbl->interfacevftbllength[i];
+               for (; mpp != mppend; ++mpp)
+                       if (*mpp == oldentrypoint) {
+                               replace_patch_method_pointer(mpp, entrypoint, "interface");
+                       }
+       }
+}
+
+
+/* replace_patch_class_hierarchy ***********************************************
+
+   Patch a method in all loaded classes.
+
+   IN:
+          m................the method to patch
+          oldentrypoint....the old entrypoint to replace
+          entrypoint.......the new entrypoint
+
+*******************************************************************************/
+
+struct replace_patch_data_t {
+       methodinfo *m;
+       u1         *oldentrypoint;
+       u1         *entrypoint;
+};
+
+void replace_patch_callback(classinfo *c, struct replace_patch_data_t *pd)
+{
+       vftbl_t *vftbl = c->vftbl;
+
+       if (vftbl != NULL
+               && vftbl->vftbllength > pd->m->vftblindex
+               && vftbl->table[pd->m->vftblindex] != &asm_abstractmethoderror
+               && code_get_methodinfo_for_pv(vftbl->table[pd->m->vftblindex]) == pd->m)
+       {
+               replace_patch_class(c->vftbl, pd->m, pd->oldentrypoint, pd->entrypoint);
+       }
+}
+
+void replace_patch_class_hierarchy(methodinfo *m,
+                                                                  u1 *oldentrypoint,
+                                                                  u1 *entrypoint)
+{
+       struct replace_patch_data_t pd;
+
+       pd.m = m;
+       pd.oldentrypoint = oldentrypoint;
+       pd.entrypoint = entrypoint;
+
+       DOLOG_SHORT( printf("patching class hierarchy: ");
+                            method_println(m); );
+
+       classcache_foreach_loaded_class(
+                       (classcache_foreach_functionptr_t) &replace_patch_callback,
+                       (void*) &pd);
 }
 
 
@@ -1408,112 +1783,85 @@ u1* replace_pop_activation_record(executionstate_t *es,
 
    IN:
           ra...............return address pointing after the call site
-          calleeframe......source frame of the callee
-          calleecode.......the codeinfo of the callee
+          callerframe......source frame of the caller
+          calleeframe......source frame of the callee, must have been mapped
 
 *******************************************************************************/
 
-void replace_patch_future_calls(u1 *ra, sourceframe_t *calleeframe, codeinfo *calleecode)
+void replace_patch_future_calls(u1 *ra,
+                                                               sourceframe_t *callerframe,
+                                                               sourceframe_t *calleeframe)
 {
-       methodptr *mpp;
-       bool       atentry;
-#if !defined(NDEBUG)
-       codeinfo  *oldcode;
-       codeinfo  *newcode;
-#endif
-#if defined(REPLACE_VERBOSE)
-       s4         i;
-       char      *logkind;
-       int        disas = 0;
-#endif
+       u1            *patchpos;
+       methodptr      entrypoint;
+       methodptr      oldentrypoint;
+       bool           atentry;
+       void          *pv;
+       codeinfo      *calleecode;
+       methodinfo    *calleem;
+       java_object_t *obj;
+       vftbl_t       *vftbl;
 
        assert(ra);
+       assert(callerframe->down == calleeframe);
+
+       /* get the new codeinfo and the method that shall be entered */
+
+       calleecode = calleeframe->tocode;
        assert(calleecode);
 
-       mpp = NULL;
+       calleem = calleeframe->method;
+       assert(calleem == calleecode->m);
+
+       entrypoint = (methodptr) calleecode->entrypoint;
+
+       /* check if we are at an method entry rplpoint at the innermost frame */
 
        atentry = (calleeframe->down == NULL)
-                       && !(calleecode->m->flags & ACC_STATIC)
-                       && (calleeframe->readrp->id == 0); /* XXX */
+                       && !(calleem->flags & ACC_STATIC)
+                       && (calleeframe->fromrp->id == 0); /* XXX */
 
-       DOLOG( printf("bytes at patch position:");
-                  for (i=0; i<16; ++i)
-                          printf(" %02x", ra[-16+i]);
-                  printf("\n"); );
+       /* get the position to patch, in case it was a statically bound call   */
 
-       if (ra[-2] == 0xff && ra[-1] == 0xd1
-                       && ra[-7] == 0xb9)
-       {
-               DOLOG_SHORT( logkind = "static   "; );
-               DOLOG( printf("PATCHING static call to "); method_println(calleecode->m); disas = 7; );
-               REPLACE_COUNT(stat_staticpatch);
-               mpp = (methodptr*)(ra - 6);
-       }
-#if defined(REPLACE_PATCH_DYNAMIC_CALL)
-       else if (ra[-2] == 0xff && ra[-1] == 0xd2
-                               && ra[-8] == 0x8b && ra[-7] == 0x91
-                               && atentry)
-       {
-               java_objectheader *obj;
-               u1 *table;
-               u4 offset;
+       pv = callerframe->fromcode->entrypoint;
+       patchpos = md_jit_method_patch_address(pv, ra, NULL);
 
-               DOLOG_SHORT( printf("\tinstance: "); java_value_print(TYPE_ADR, calleeframe->instance);
-                                        printf("\n"); );
+       if (patchpos == NULL) {
+               /* the call was dispatched dynamically */
 
-               assert(calleeframe->instance != 0);
+               /* we can only patch such calls if we are at the entry point */
 
-               obj = (java_objectheader *) (ptrint) calleeframe->instance;
-               table = (u1*) obj->vftbl;
-               offset = *(u4*)(ra - 6);
+               if (!atentry)
+                       return;
 
-               if (ra[-10] == 0x8b && ra[-9] == 0x08) {
-                       mpp = (methodptr *) (table + offset);
-                       DOLOG_SHORT( logkind = "virtual  "; );
-                       DOLOG( printf("updating virtual call at %p\n", (void*) ra); disas = 8);
-               }
-               else {
-                       u4 ioffset = *(u4*)(ra - 12);
-                       u1 *itable = *(u1**)(table + ioffset);
+               assert((calleem->flags & ACC_STATIC) == 0);
 
-                       assert(ra[-14] == 0x8b && ra[-13] == 0x89);
-                       mpp = (methodptr *) (itable + offset);
-                       DOLOG_SHORT( logkind = "interface"; );
-                       DOLOG( printf("updating interface call at %p\n", (void*) ra); disas = 14);
-               }
-       }
-#endif /* defined(REPLACE_PATCH_DYNAMIC_CALL) */
+               oldentrypoint = calleeframe->fromcode->entrypoint;
 
-       if (mpp == NULL)
-               return;
+               /* we need to know the instance */
 
-       DOLOG(
-               u1* u1ptr = ra - disas;
-               DISASSINSTR(u1ptr);
-               DISASSINSTR(u1ptr);
-               if (disas > 8)
-                       DISASSINSTR(u1ptr);
-               fflush(stdout);
-       );
+               if (!calleeframe->instance.a) {
+                       DOLOG_SHORT( printf("WARNING: object instance unknown!\n"); );
+                       replace_patch_class_hierarchy(calleem, oldentrypoint, entrypoint);
+                       return;
+               }
 
-       DOLOG( printf("patch method pointer from: %p to %p\n",
-                                 (void*) *mpp, (void*)calleecode->entrypoint); );
+               /* get the vftbl */
 
-#if !defined(NDEBUG)
-       oldcode = *(codeinfo **)((u1*)(*mpp) + CodeinfoPointer);
-       newcode = *(codeinfo **)((u1*)(calleecode->entrypoint) + CodeinfoPointer);
+               obj = calleeframe->instance.a;
+               vftbl = obj->vftbl;
 
-       DOLOG_SHORT( printf("\tpatch %s %p ", logkind, (void*) oldcode->entrypoint);
-                                method_println(oldcode->m);
-                                printf("\t      with      %p ", (void*) newcode->entrypoint);
-                                method_println(newcode->m); );
+               assert(vftbl->class->vftbl == vftbl);
 
-       assert(oldcode->m == newcode->m);
-#endif
+               DOLOG_SHORT( printf("\tclass: "); class_println(vftbl->class); );
 
-       /* write the new entrypoint */
+               replace_patch_class(vftbl, calleem, oldentrypoint, entrypoint);
+       }
+       else {
+               /* the call was statically bound */
 
-       *mpp = (methodptr) calleecode->entrypoint;
+               replace_patch_method_pointer((methodptr *) patchpos, entrypoint, "static   ");
+       }
 }
 
 
@@ -1527,9 +1875,9 @@ void replace_patch_future_calls(u1 *ra, sourceframe_t *calleeframe, codeinfo *ca
    IN:
           es...............execution state
           rpcall...........the replacement point at the call site
-          callerframe......source frame of the caller
-          calleecode.......the codeinfo of the callee
-          calleeframe......source frame of the callee
+          callerframe......source frame of the caller, or NULL for creating the
+                           first frame
+          calleeframe......source frame of the callee, must have been mapped
 
    OUT:
        *es..............the execution state after pushing the stack frame
@@ -1539,7 +1887,6 @@ void replace_patch_future_calls(u1 *ra, sourceframe_t *calleeframe, codeinfo *ca
 void replace_push_activation_record(executionstate_t *es,
                                                                        rplpoint *rpcall,
                                                                        sourceframe_t *callerframe,
-                                                                       codeinfo *calleecode,
                                                                        sourceframe_t *calleeframe)
 {
        s4           reg;
@@ -1548,62 +1895,94 @@ void replace_push_activation_record(executionstate_t *es,
        stackslot_t *basesp;
        stackslot_t *sp;
        u1          *ra;
+       codeinfo    *calleecode;
 
        assert(es);
-       assert(rpcall && rpcall->type == RPLPOINT_TYPE_CALL);
-       assert(calleecode);
-       assert(callerframe);
+       assert(!rpcall || callerframe);
+    assert(!rpcall || rpcall->type == RPLPOINT_TYPE_CALL);
+       assert(!rpcall || rpcall == callerframe->torp);
        assert(calleeframe);
-       assert(calleeframe == callerframe->down);
+       assert(!callerframe || calleeframe == callerframe->down);
 
-       /* write the return address */
+       /* the compilation unit we are entering */
 
-       es->sp -= SIZE_OF_STACKSLOT;
+       calleecode = calleeframe->tocode;
+       assert(calleecode);
+
+       /* calculate the return address */
 
-       ra = rpcall->pc + rpcall->callsize;
+       if (rpcall)
+               ra = rpcall->pc + rpcall->callsize;
+       else
+               ra = es->pc + 1 /* XXX this is ugly */;
 
-       DOLOG( printf("writing return address %p to %p\n",
-                               (void*) ra, (void*) es->sp); );
+       /* write the return address */
+
+#if defined(REPLACE_RA_BETWEEN_FRAMES)
+       es->sp -= SIZE_OF_STACKSLOT;
 
        *((stackslot_t *)es->sp) = (stackslot_t) ra;
+#endif /* REPLACE_RA_BETWEEN_FRAMES */
+
+#if defined(REPLACE_REG_RA)
+       es->intregs[REPLACE_REG_RA] = (ptrint) ra;
+#endif
 
        /* we move into a new code unit */
 
        es->code = calleecode;
 
-       /* set the new pc XXX not needed */
+       /* set the new pc XXX not needed? */
 
-       es->pc = es->code->entrypoint;
+       es->pc = calleecode->entrypoint;
 
        /* build the stackframe */
 
        DOLOG( printf("building stackframe of %d words at %p\n",
-                               es->code->stackframesize, (void*)es->sp); );
+                                 calleecode->stackframesize, (void*)es->sp); );
 
        sp = (stackslot_t *) es->sp;
        basesp = sp;
 
-       sp -= es->code->stackframesize;
+       sp -= calleecode->stackframesize;
        es->sp = (u1*) sp;
 
        /* in debug mode, invalidate stack frame first */
 
-#if !defined(NDEBUG)
+       /* XXX may not invalidate linkage area used by native code! */
+#if !defined(NDEBUG) && 0
        for (i=0; i<(basesp - sp); ++i) {
                sp[i] = 0xdeaddeadU;
        }
 #endif
 
+       /* save the return address register */
+
+#if defined(REPLACE_RA_TOP_OF_FRAME)
+#if defined(REPLACE_LEAFMETHODS_RA_REGISTER)
+       if (!code_is_leafmethod(calleecode))
+#endif
+               *--basesp = (ptrint) ra;
+#endif /* REPLACE_RA_TOP_OF_FRAME */
+
+#if defined(REPLACE_RA_LINKAGE_AREA)
+#if defined(REPLACE_LEAFMETHODS_RA_REGISTER)
+       if (!code_is_leafmethod(calleecode))
+#endif
+               basesp[LA_LR_OFFSET / sizeof(stackslot_t)] = (ptrint) ra;
+#endif /* REPLACE_RA_LINKAGE_AREA */
+
        /* save int registers */
 
        reg = INT_REG_CNT;
-       for (i=0; i<es->code->savedintcount; ++i) {
+       for (i=0; i<calleecode->savedintcount; ++i) {
                while (nregdescint[--reg] != REG_SAV)
                        ;
                *--basesp = es->intregs[reg];
 
-#if !defined(NDEBUG)
-               es->intregs[reg] = 0x44dead4444dead44ULL;
+               /* XXX may not clobber saved regs used by native code! */
+#if !defined(NDEBUG) && 0
+               es->intregs[reg] = (ptrint) 0x44dead4444dead44ULL;
 #endif
        }
 
@@ -1611,37 +1990,56 @@ void replace_push_activation_record(executionstate_t *es,
 
        /* XXX align? */
        reg = FLT_REG_CNT;
-       for (i=0; i<es->code->savedfltcount; ++i) {
+       for (i=0; i<calleecode->savedfltcount; ++i) {
                while (nregdescfloat[--reg] != REG_SAV)
                        ;
                basesp -= STACK_SLOTS_PER_FLOAT;
-               *(u8*)basesp = es->fltregs[reg];
+               *(double*)basesp = es->fltregs[reg];
 
-#if !defined(NDEBUG)
-               es->fltregs[reg] = 0x44dead4444dead44ULL;
+               /* XXX may not clobber saved regs used by native code! */
+#if !defined(NDEBUG) && 0
+               *(u8*)&(es->fltregs[reg]) = 0x44dead4444dead44ULL;
+#endif
+       }
+
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+       /* save adr registers */
+
+       reg = ADR_REG_CNT;
+       for (i=0; i<calleecode->savedadrcount; ++i) {
+               while (nregdescadr[--reg] != REG_SAV)
+                       ;
+               *--basesp = es->adrregs[reg];
+
+               /* XXX may not clobber saved regs used by native code! */
+#if !defined(NDEBUG) && 0
+               es->adrregs[reg] = (ptrint) 0x44dead4444dead44ULL;
 #endif
        }
+#endif
 
        /* write slots used for synchronization */
 
-       count = code_get_sync_slot_count(es->code);
+       count = code_get_sync_slot_count(calleecode);
        assert(count == calleeframe->syncslotcount);
        for (i=0; i<count; ++i) {
-               sp[es->code->memuse + i] = calleeframe->syncslots[i];
+               sp[calleecode->memuse + i] = calleeframe->syncslots[i].p;
        }
 
        /* set the PV */
 
-       es->pv = es->code->entrypoint;
+       es->pv = calleecode->entrypoint;
 
        /* redirect future invocations */
 
+       if (callerframe && rpcall) {
 #if defined(REPLACE_PATCH_ALL)
-       if (rpcall->type == callerframe->readrp->type)
+               if (rpcall->type == callerframe->fromrp->type)
 #else
-       if (rpcall == callerframe->readrp)
+               if (rpcall == callerframe->fromrp)
 #endif
-               replace_patch_future_calls(ra, calleeframe, calleecode);
+                       replace_patch_future_calls(ra, callerframe, calleeframe);
+       }
 }
 
 
@@ -1652,7 +2050,7 @@ void replace_push_activation_record(executionstate_t *es,
    
    IN:
           code.............the codeinfo in which to search the rplpoint
-          ss...............the source state defining the position to look for
+          frame............the source frame defining the position to look for
           parent...........parent replacement point to match
 
    RETURN VALUE:
@@ -1661,10 +2059,9 @@ void replace_push_activation_record(executionstate_t *es,
 *******************************************************************************/
 
 rplpoint * replace_find_replacement_point(codeinfo *code,
-                                                                                 sourcestate_t *ss,
+                                                                                 sourceframe_t *frame,
                                                                                  rplpoint *parent)
 {
-       sourceframe_t *frame;
        methodinfo *m;
        rplpoint *rp;
        s4        i;
@@ -1672,9 +2069,7 @@ rplpoint * replace_find_replacement_point(codeinfo *code,
        s4        stacki;
        rplalloc *ra;
 
-       assert(ss);
-
-       frame = ss->frames;
+       assert(code);
        assert(frame);
 
        DOLOG( printf("searching replacement point for:\n");
@@ -1701,13 +2096,13 @@ rplpoint * replace_find_replacement_point(codeinfo *code,
                                if (ra->type == TYPE_RET) {
                                        if (ra->index == RPLALLOC_STACK) {
                                                assert(stacki < frame->javastackdepth);
-                                               if (frame->javastack[stacki] != ra->regoff)
+                                               if (frame->javastack[stacki].i != ra->regoff)
                                                        goto no_match;
                                                stacki++;
                                        }
                                        else {
                                                assert(ra->index >= 0 && ra->index < frame->javalocalcount);
-                                               if (frame->javalocals[ra->index] != ra->regoff)
+                                               if (frame->javalocals[ra->index].i != ra->regoff)
                                                        goto no_match;
                                        }
                                }
@@ -1729,52 +2124,652 @@ no_match:
        }
 #endif
 
-       vm_abort("no matching replacement point found");
-       return NULL; /* NOT REACHED */
+       vm_abort("no matching replacement point found");
+       return NULL; /* NOT REACHED */
+}
+
+
+/* replace_find_replacement_point_for_pc ***************************************
+
+   Find the nearest replacement point at or before the given PC. The
+   given PC has to be between (rp->pc) and (rp->pc+rp->callsize) for
+   the replacement point to be found.
+
+   IN:
+       code.............compilation unit the PC is in
+          pc...............the machine code PC
+
+   RETURN VALUE:
+       the replacement point found, or
+          NULL if no replacement point was found
+
+*******************************************************************************/
+
+rplpoint *replace_find_replacement_point_for_pc(codeinfo *code, u1 *pc)
+{
+       rplpoint *found;
+       rplpoint *rp;
+       s4        i;
+
+       DOLOG( printf("searching for rp at pc:%p in %p ", (void*)pc, (void*)code);
+                  method_println(code->m); );
+
+       found = NULL;
+
+       rp = code->rplpoints;
+       for (i=0; i<code->rplpointcount; ++i, ++rp) {
+               DOLOG( replace_replacement_point_println(rp, 2); );
+               if (rp->pc <= pc && rp->pc + rp->callsize >= pc)
+                       found = rp;
+       }
+
+       return found;
+}
+
+
+/* replace_pop_native_frame ****************************************************
+
+   Unroll a native frame in the execution state and create a source frame
+   for it.
+
+   IN:
+          es...............current execution state
+          ss...............the current source state
+          sfi..............stackframeinfo for the native frame
+
+   OUT:
+       es...............execution state after unrolling the native frame
+          ss...............gets the added native source frame
+
+*******************************************************************************/
+
+static void replace_pop_native_frame(executionstate_t *es,
+                                                                        sourcestate_t *ss,
+                                                                        stackframeinfo_t *sfi)
+{
+       sourceframe_t *frame;
+       codeinfo      *code;
+       s4             i,j;
+
+       assert(sfi);
+
+       frame = replace_new_sourceframe(ss);
+
+       frame->sfi = sfi;
+
+       /* remember pc and size of native frame */
+
+       frame->nativepc = es->pc;
+       frame->nativeframesize = (es->sp != 0) ? (sfi->sp - es->sp) : 0;
+       assert(frame->nativeframesize >= 0);
+
+       /* remember values of saved registers */
+
+       j = 0;
+       for (i=0; i<INT_REG_CNT; ++i) {
+               if (nregdescint[i] == REG_SAV)
+                       frame->nativesavint[j++] = es->intregs[i];
+       }
+
+       j = 0;
+       for (i=0; i<FLT_REG_CNT; ++i) {
+               if (nregdescfloat[i] == REG_SAV)
+                       frame->nativesavflt[j++] = es->fltregs[i];
+       }
+
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+       j = 0;
+       for (i=0; i<ADR_REG_CNT; ++i) {
+               if (nregdescadr[i] == REG_SAV)
+                       frame->nativesavadr[j++] = es->adrregs[i];
+       }
+#endif
+
+       /* restore saved registers */
+
+#if defined(ENABLE_GC_CACAO) && !defined(HAS_ADDRESS_REGISTER_FILE)
+       j = 0;
+       for (i=0; i<INT_REG_CNT; ++i) {
+               if (nregdescint[i] == REG_SAV)
+                       es->intregs[i] = sfi->intregs[j++];
+       }
+#else
+       /* XXX we don't have them, yet, in the sfi, so clear them */
+
+       for (i=0; i<INT_REG_CNT; ++i) {
+               if (nregdescint[i] == REG_SAV)
+                       es->intregs[i] = 0;
+       }
+#endif
+
+       /* XXX we don't have float registers in the sfi, so clear them */
+
+       for (i=0; i<FLT_REG_CNT; ++i) {
+               if (nregdescfloat[i] == REG_SAV)
+                       es->fltregs[i] = 0.0;
+       }
+
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+# if defined(ENABLE_GC_CACAO)
+       j = 0;
+       for (i=0; i<ADR_REG_CNT; ++i) {
+               if (nregdescadr[i] == REG_SAV)
+                       es->adrregs[i] = sfi->adrregs[j++];
+       }
+# else
+       for (i=0; i<ADR_REG_CNT; ++i) {
+               if (nregdescadr[i] == REG_SAV)
+                       es->adrregs[i] = 0;
+       }
+# endif
+#endif
+
+       /* restore codeinfo of the native stub */
+
+       code = code_get_codeinfo_for_pv(sfi->pv);
+
+       /* restore sp, pv, pc and codeinfo of the parent method */
+
+       /* XXX michi: use this instead:
+       es->sp = sfi->sp + code->stackframesize; */
+       es->sp   = sfi->sp + (*(s4 *) (sfi->pv + FrameSize));
+#if defined(REPLACE_RA_BETWEEN_FRAMES)
+       es->sp  += SIZE_OF_STACKSLOT; /* skip return address */
+#endif
+       es->pv   = md_codegen_get_pv_from_pc(sfi->ra);
+       es->pc   = ((sfi->xpc) ? sfi->xpc : sfi->ra) - 1;
+       es->code = code_get_codeinfo_for_pv(es->pv);
+}
+
+
+/* replace_push_native_frame ***************************************************
+
+   Rebuild a native frame onto the execution state and remove its source frame.
+
+   Note: The native frame is "rebuild" by setting fields like PC and stack
+         pointer in the execution state accordingly. Values in the
+                stackframeinfo may be modified, but the actual stack frame of the
+                native code is not touched.
+
+   IN:
+          es...............current execution state
+          ss...............the current source state
+
+   OUT:
+       es...............execution state after re-rolling the native frame
+          ss...............the native source frame is removed
+
+*******************************************************************************/
+
+static void replace_push_native_frame(executionstate_t *es, sourcestate_t *ss)
+{
+       sourceframe_t *frame;
+       s4             i,j;
+
+       assert(es);
+       assert(ss);
+
+       DOLOG( printf("pushing native frame\n"); );
+
+       /* remove the frame from the source state */
+
+       frame = ss->frames;
+       assert(frame);
+       assert(REPLACE_IS_NATIVE_FRAME(frame));
+
+       ss->frames = frame->down;
+
+       /* skip sp for the native stub */
+
+       es->sp -= (*(s4 *) (frame->sfi->pv + FrameSize));
+#if defined(REPLACE_RA_BETWEEN_FRAMES)
+       es->sp -= SIZE_OF_STACKSLOT; /* skip return address */
+#endif
+
+       /* assert that the native frame has not moved */
+
+       assert(es->sp == frame->sfi->sp);
+
+       /* update saved registers in the stackframeinfo */
+
+#if defined(ENABLE_GC_CACAO)
+       j = 0;
+# if !defined(HAS_ADDRESS_REGISTER_FILE)
+       for (i=0; i<INT_REG_CNT; ++i) {
+               if (nregdescint[i] == REG_SAV)
+                       frame->sfi->intregs[j++] = es->intregs[i];
+       }
+# else
+       for (i=0; i<ADR_REG_CNT; ++i) {
+               if (nregdescadr[i] == REG_SAV)
+                       frame->sfi->adrregs[j++] = es->adrregs[i];
+       }
+# endif
+
+       /* XXX leave float registers untouched here */
+#endif
+
+       /* restore saved registers */
+
+       j = 0;
+       for (i=0; i<INT_REG_CNT; ++i) {
+               if (nregdescint[i] == REG_SAV)
+                       es->intregs[i] = frame->nativesavint[j++];
+       }
+
+       j = 0;
+       for (i=0; i<FLT_REG_CNT; ++i) {
+               if (nregdescfloat[i] == REG_SAV)
+                       es->fltregs[i] = frame->nativesavflt[j++];
+       }
+
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+       j = 0;
+       for (i=0; i<ADR_REG_CNT; ++i) {
+               if (nregdescadr[i] == REG_SAV)
+                       es->adrregs[i] = frame->nativesavadr[j++];
+       }
+#endif
+
+       /* skip the native frame on the machine stack */
+
+       es->sp -= frame->nativeframesize;
+
+       /* set the pc the next frame must return to */
+
+       es->pc = frame->nativepc;
+}
+
+
+/* replace_recover_source_state ************************************************
+
+   Recover the source state from the given replacement point and execution
+   state.
+
+   IN:
+       rp...............replacement point that has been reached, if any
+          sfi..............stackframeinfo, if called from native code
+          es...............execution state at the replacement point rp
+
+   RETURN VALUE:
+       the source state
+
+*******************************************************************************/
+
+sourcestate_t *replace_recover_source_state(rplpoint *rp,
+                                                                                       stackframeinfo_t *sfi,
+                                                                                   executionstate_t *es)
+{
+       sourcestate_t *ss;
+       u1            *ra;
+       bool           locked;
+#if defined(REPLACE_STATISTICS)
+       s4             depth;
+#endif
+
+       /* create the source frame structure in dump memory */
+
+       ss = DNEW(sourcestate_t);
+       ss->frames = NULL;
+
+       /* each iteration of the loop recovers one source frame */
+
+       depth = 0;
+       locked = false;
+
+       while (rp || sfi) {
+
+               DOLOG( replace_executionstate_println(es); );
+
+               /* if we are not at a replacement point, it is a native frame */
+
+               if (rp == NULL) {
+                       DOLOG( printf("native frame: sfi: "); replace_stackframeinfo_println(sfi); );
+
+                       locked = true;
+                       replace_pop_native_frame(es, ss, sfi);
+                       sfi = sfi->prev;
+
+                       if (es->code == NULL)
+                               continue;
+
+                       goto after_machine_frame;
+               }
+
+               /* read the values for this source frame from the execution state */
+
+               DOLOG( printf("recovering source state for%s:\n",
+                                       (ss->frames == NULL) ? " TOPFRAME" : "");
+                          replace_replacement_point_println(rp, 1); );
+
+               replace_read_executionstate(rp, es, ss, ss->frames == NULL);
+
+#if defined(ENABLE_VMLOG)
+               vmlog_cacao_unrol_method(ss->frames->method);
+#endif
+
+#if defined(REPLACE_STATISTICS)
+               REPLACE_COUNT(stat_frames);
+               depth++;
+               replace_statistics_source_frame(ss->frames);
+#endif
+
+               /* in locked areas (below native frames), identity map the frame */
+
+               if (locked) {
+                       ss->frames->torp = ss->frames->fromrp;
+                       ss->frames->tocode = ss->frames->fromcode;
+               }
+
+               /* unroll to the next (outer) frame */
+
+               if (rp->parent) {
+                       /* this frame is in inlined code */
+
+                       DOLOG( printf("INLINED!\n"); );
+
+                       rp = rp->parent;
+
+                       assert(rp->type == RPLPOINT_TYPE_INLINE);
+                       REPLACE_COUNT(stat_unroll_inline);
+               }
+               else {
+                       /* this frame had been called at machine-level. pop it. */
+
+                       DOLOG( printf("UNWIND\n"); );
+
+                       ra = replace_pop_activation_record(es, ss->frames);
+                       if (ra == NULL) {
+                               DOLOG( printf("REACHED NATIVE CODE\n"); );
+
+                               rp = NULL;
+
+#if !defined(ENABLE_GC_CACAO)
+                               break; /* XXX remove to activate native frames */
+#endif
+                               continue;
+                       }
+
+                       /* find the replacement point at the call site */
+
+after_machine_frame:
+                       rp = replace_find_replacement_point_for_pc(es->code, es->pc);
+
+                       if (rp == NULL)
+                               vm_abort("could not find replacement point while unrolling call");
+
+                       DOLOG( printf("found replacement point.\n");
+                                       replace_replacement_point_println(rp, 1); );
+
+                       assert(rp->type == RPLPOINT_TYPE_CALL);
+                       REPLACE_COUNT(stat_unroll_call);
+               }
+       } /* end loop over source frames */
+
+       REPLACE_COUNT_DIST(stat_dist_frames, depth);
+
+       return ss;
+}
+
+
+/* replace_map_source_state ****************************************************
+
+   Map each source frame in the given source state to a target replacement
+   point and compilation unit. If no valid code is available for a source
+   frame, it is (re)compiled.
+
+   IN:
+       ss...............the source state
+
+   OUT:
+       ss...............the source state, modified: The `torp` and `tocode`
+                           fields of each source frame are set.
+
+   RETURN VALUE:
+       true.............everything went ok
+          false............an exception has been thrown
+
+*******************************************************************************/
+
+static bool replace_map_source_state(sourcestate_t *ss)
+{
+       sourceframe_t *frame;
+       codeinfo      *code;
+       rplpoint      *rp;
+       rplpoint      *parent; /* parent of inlined rplpoint */
+#if defined(REPLACE_STATISTICS)
+       codeinfo      *oldcode;
+#endif
+
+       parent = NULL;
+       code = NULL;
+
+       /* iterate over the source frames from outermost to innermost */
+
+       for (frame = ss->frames; frame != NULL; frame = frame->down) {
+
+               /* XXX skip native frames */
+
+               if (REPLACE_IS_NATIVE_FRAME(frame)) {
+                       parent = NULL;
+                       continue;
+               }
+
+               /* map frames which are not already mapped */
+
+               if (frame->tocode) {
+                       code = frame->tocode;
+                       rp = frame->torp;
+                       assert(rp);
+               }
+               else {
+                       assert(frame->torp == NULL);
+
+                       if (parent == NULL) {
+                               /* find code for this frame */
+
+#if defined(REPLACE_STATISTICS)
+                               oldcode = frame->method->code;
+#endif
+                               /* request optimization of hot methods and their callers */
+
+                               if (frame->method->hitcountdown < 0
+                                       || (frame->down && frame->down->method->hitcountdown < 0))
+                                       jit_request_optimization(frame->method);
+
+                               code = jit_get_current_code(frame->method);
+
+                               if (code == NULL)
+                                       return false; /* exception */
+
+                               REPLACE_COUNT_IF(stat_recompile, code != oldcode);
+                       }
+
+                       assert(code);
+
+                       /* map this frame */
+
+                       rp = replace_find_replacement_point(code, frame, parent);
+
+                       frame->tocode = code;
+                       frame->torp = rp;
+               }
+
+               if (rp->type == RPLPOINT_TYPE_CALL) {
+                       parent = NULL;
+               }
+               else {
+                       /* inlining */
+                       parent = rp;
+               }
+       }
+
+       return true;
+}
+
+
+/* replace_map_source_state_identity *******************************************
+
+   Map each source frame in the given source state to the same replacement
+   point and compilation unit it was derived from. This is mainly used for
+   garbage collection.
+
+   IN:
+       ss...............the source state
+
+   OUT:
+       ss...............the source state, modified: The `torp` and `tocode`
+                           fields of each source frame are set.
+
+*******************************************************************************/
+
+#if defined(ENABLE_GC_CACAO)
+static void replace_map_source_state_identity(sourcestate_t *ss)
+{
+       sourceframe_t *frame;
+
+       /* iterate over the source frames from outermost to innermost */
+
+       for (frame = ss->frames; frame != NULL; frame = frame->down) {
+
+               /* skip native frames */
+
+               if (REPLACE_IS_NATIVE_FRAME(frame)) {
+                       continue;
+               }
+
+               /* map frames using the identity mapping */
+
+               if (frame->tocode) {
+                       assert(frame->tocode == frame->fromcode);
+                       assert(frame->torp   == frame->fromrp);
+               } else {
+                       assert(frame->tocode == NULL);
+                       assert(frame->torp   == NULL);
+                       frame->tocode = frame->fromcode;
+                       frame->torp   = frame->fromrp;
+               }
+       }
+}
+#endif
+
+
+/* replace_build_execution_state ***********************************************
+
+   Build an execution state for the given (mapped) source state.
+
+   !!! CAUTION: This function rewrites the machine stack !!!
+
+   THIS FUNCTION MUST BE CALLED USING A SAFE STACK AREA!
+
+   IN:
+       ss...............the source state. Must have been mapped by
+                                               replace_map_source_state before.
+          es...............the base execution state on which to build
+
+   OUT:
+       *es..............the new execution state
+
+*******************************************************************************/
+
+static void replace_build_execution_state(sourcestate_t *ss,
+                                                                                 executionstate_t *es)
+{
+       rplpoint      *rp;
+       sourceframe_t *prevframe;
+       rplpoint      *parent;
+
+       parent = NULL;
+       prevframe = NULL;
+       rp = NULL;
+
+       while (ss->frames) {
+
+               if (REPLACE_IS_NATIVE_FRAME(ss->frames)) {
+                       prevframe = ss->frames;
+                       replace_push_native_frame(es, ss);
+                       parent = NULL;
+                       rp = NULL;
+                       continue;
+               }
+
+               if (parent == NULL) {
+                       /* create a machine-level stack frame */
+
+                       DOLOG( printf("pushing activation record for:\n");
+                                  if (rp) replace_replacement_point_println(rp, 1);
+                                  else printf("\tfirst frame\n"); );
+
+                       replace_push_activation_record(es, rp, prevframe, ss->frames);
+
+                       DOLOG( replace_executionstate_println(es); );
+               }
+
+               rp = ss->frames->torp;
+               assert(rp);
+
+               DOLOG( printf("creating execution state for%s:\n",
+                               (ss->frames->down == NULL) ? " TOPFRAME" : "");
+                          replace_replacement_point_println(ss->frames->fromrp, 1);
+                          replace_replacement_point_println(rp, 1); );
+
+               es->code = ss->frames->tocode;
+               prevframe = ss->frames;
+
+#if defined(ENABLE_VMLOG)
+               vmlog_cacao_rerol_method(ss->frames->method);
+#endif
+
+               replace_write_executionstate(rp, es, ss, ss->frames->down == NULL);
+
+               DOLOG( replace_executionstate_println(es); );
+
+               if (rp->type == RPLPOINT_TYPE_CALL) {
+                       parent = NULL;
+               }
+               else {
+                       /* inlining */
+                       parent = rp;
+               }
+       }
 }
 
 
 /* replace_me ******************************************************************
  
-   This function is called by asm_replacement_out when a thread reaches
+   This function is called by the signal handler when a thread reaches
    a replacement point. `replace_me` must map the execution state to the
    target replacement point and let execution continue there.
 
-   This function never returns!
-  
+   THIS FUNCTION MUST BE CALLED USING A SAFE STACK AREA!
+
    IN:
        rp...............replacement point that has been reached
-          es...............execution state read by asm_replacement_out
+       es...............execution state read by signal handler
   
 *******************************************************************************/
 
-void replace_me(rplpoint *rp, executionstate_t *es)
+static void replace_me(rplpoint *rp, executionstate_t *es)
 {
-       sourcestate_t  ss;
-       s4             dumpsize;
-       rplpoint      *candidate;
-       codeinfo      *code;
-       s4             i;
-       s4             depth;
-       rplpoint      *origrp;
-       rplpoint      *parent;
-       u1            *ra;
-       sourceframe_t *prevframe;
-#if defined(REPLACE_STATISTICS)
-       codeinfo      *oldcode;
+       stackframeinfo_t    *sfi;
+       sourcestate_t       *ss;
+       sourceframe_t       *frame;
+       s4                   dumpsize;
+       codeinfo            *origcode;
+       rplpoint            *origrp;
+#if defined(ENABLE_THREADS) && defined(ENABLE_GC_CACAO)
+       threadobject        *thread;
 #endif
 
-       origrp = rp;
-       es->code = rp->code;
+       origcode = es->code;
+       origrp   = rp;
 
        DOLOG_SHORT( printf("REPLACING(%d %p): (id %d %p) ",
                                 stat_replacements, (void*)THREADOBJECT,
                                 rp->id, (void*)rp);
                                 method_println(es->code->m); );
 
-       DOLOG( printf("replace_me(%p,%p)\n",(void*)rp,(void*)es); fflush(stdout);
-                  replace_replacement_point_println(rp, 1);
-                  replace_executionstate_println(es); );
+       DOLOG( replace_replacement_point_println(rp, 1); );
 
        REPLACE_COUNT(stat_replacements);
 
@@ -1782,166 +2777,322 @@ void replace_me(rplpoint *rp, executionstate_t *es)
 
        dumpsize = dump_size();
 
-       /* read execution state of old code */
-
-       ss.frames = NULL;
+       /* get the stackframeinfo for the current thread */
 
-       /* XXX testing */
+       sfi = STACKFRAMEINFO;
 
-       depth = 0;
-       candidate = rp;
-       do {
-               DOLOG( printf("recovering source state for%s:\n",
-                                       (ss.frames == NULL) ? " TOPFRAME" : "");
-                          replace_replacement_point_println(candidate, 1); );
+       /* recover source state */
 
-               replace_read_executionstate(candidate, es, &ss, ss.frames == NULL);
-               REPLACE_COUNT(stat_frames);
-               depth++;
+       ss = replace_recover_source_state(rp, sfi, es);
 
-#if defined(REPLACE_STATISTICS)
-               {
-               int adr = 0; int ret = 0; int prim = 0; int vd = 0; int n = 0;
-               for (i=0; i<ss.frames->javalocalcount; ++i) {
-                       switch (ss.frames->javalocaltype[i]) {
-                               case TYPE_ADR: adr++; break;
-                               case TYPE_RET: ret++; break;
-                               case TYPE_INT: case TYPE_LNG: case TYPE_FLT: case TYPE_DBL: prim++; break;
-                               case TYPE_VOID: vd++; break;
-                               default: assert(0);
-                       }
-                       n++;
-               }
-               REPLACE_COUNT_DIST(stat_dist_locals, n);
-               REPLACE_COUNT_DIST(stat_dist_locals_adr, adr);
-               REPLACE_COUNT_DIST(stat_dist_locals_void, vd);
-               REPLACE_COUNT_DIST(stat_dist_locals_ret, ret);
-               REPLACE_COUNT_DIST(stat_dist_locals_prim, prim);
-               adr = ret = prim = n = 0;
-               for (i=0; i<ss.frames->javastackdepth; ++i) {
-                       switch (ss.frames->javastacktype[i]) {
-                               case TYPE_ADR: adr++; break;
-                               case TYPE_RET: ret++; break;
-                               case TYPE_INT: case TYPE_LNG: case TYPE_FLT: case TYPE_DBL: prim++; break;
-                       }
-                       n++;
-               }
-               REPLACE_COUNT_DIST(stat_dist_stack, n);
-               REPLACE_COUNT_DIST(stat_dist_stack_adr, adr);
-               REPLACE_COUNT_DIST(stat_dist_stack_ret, ret);
-               REPLACE_COUNT_DIST(stat_dist_stack_prim, prim);
-               }
-#endif /* defined(REPLACE_STATISTICS) */
+#if defined(ENABLE_THREADS) && defined(ENABLE_GC_CACAO)
+       /* if there is a collection pending, we assume the replacement point should
+          suspend this thread */
 
-               if (candidate->parent) {
-                       DOLOG( printf("INLINED!\n"); );
-                       candidate = candidate->parent;
-                       assert(candidate->type == RPLPOINT_TYPE_INLINE);
-                       REPLACE_COUNT(stat_unroll_inline);
-               }
-               else {
-                       DOLOG( printf("UNWIND\n"); );
-                       REPLACE_COUNT(stat_unroll_call);
-                       code = es->code;
-                       ra = replace_pop_activation_record(es, ss.frames);
-                       if (ra == NULL) {
-                               DOLOG( printf("BREAKING\n"); );
-                               break;
-                       }
-                       DOLOG( replace_executionstate_println(es); );
-                       candidate = NULL;
-                       rp = es->code->rplpoints;
-                       for (i=0; i<es->code->rplpointcount; ++i, ++rp)
-                               if (rp->pc <= es->pc)
-                                       candidate = rp;
-                       if (!candidate)
-                               DOLOG( printf("NO CANDIDATE!\n"); );
-                       else {
-                               DOLOG( printf("found replacement point.\n");
-                                          replace_replacement_point_println(candidate, 1); );
-                               assert(candidate->type == RPLPOINT_TYPE_CALL);
-                       }
-               }
-       } while (candidate);
+       if (gc_pending) {
 
-       REPLACE_COUNT_DIST(stat_dist_frames, depth);
-       DOLOG( replace_sourcestate_println(&ss); );
+               thread = THREADOBJECT;
 
-       /* write execution state of new code */
+               DOLOG_SHORT( printf("REPLACEMENT: Suspending thread for GC now!\n"); );
 
-       DOLOG( replace_executionstate_println(es); );
+               /* map the sourcestate using the identity mapping */
+               replace_map_source_state_identity(ss);
 
-       code = es->code;
+               /* since we enter the same method again, we turn off rps now */
+               /* XXX michi: can we really do this? what if the rp was active before
+                  we activated it for the gc? */
+               replace_deactivate_replacement_points(origcode);
 
-       /* XXX get new code */
+               /* remember executionstate and sourcestate for this thread */
+               GC_EXECUTIONSTATE = es;
+               GC_SOURCESTATE    = ss;
 
-       parent = NULL;
+               /* really suspend this thread now (PC = 0) */
+               threads_suspend_ack(NULL, NULL);
 
-       while (ss.frames) {
+               DOLOG_SHORT( printf("REPLACEMENT: Resuming thread after GC now!\n"); );
 
-               candidate = replace_find_replacement_point(code, &ss, parent);
+       } else {
+#endif /*defined(ENABLE_THREADS) && defined(ENABLE_GC_CACAO)*/
 
-               DOLOG( printf("creating execution state for%s:\n",
-                               (ss.frames->down == NULL) ? " TOPFRAME" : "");
-                          replace_replacement_point_println(ss.frames->readrp, 1);
-                          replace_replacement_point_println(candidate, 1); );
-
-               DOLOG_SHORT( printf("\t%c%s ",
-                                        (candidate == ss.frames->readrp) ? '=' : '+',
-                                        replace_type_str[candidate->type]); 
-                                    method_println(ss.frames->method); );
-
-               prevframe = ss.frames;
-               replace_write_executionstate(candidate, es, &ss, ss.frames->down == NULL);
-               if (ss.frames == NULL)
-                       break;
-               DOLOG( replace_executionstate_println(es); );
+       /* map the source state */
 
-               if (candidate->type == RPLPOINT_TYPE_CALL) {
-#if defined(REPLACE_STATISTICS)
-                       oldcode = ss.frames->method->code;
-#endif
-                       code = jit_get_current_code(ss.frames->method);
+       if (!replace_map_source_state(ss))
+               vm_abort("exception during method replacement");
 
-                       REPLACE_COUNT_IF(stat_recompile, code != oldcode);
+       DOLOG( replace_sourcestate_println(ss); );
 
-                       DOLOG( printf("pushing activation record for:\n");
-                                  replace_replacement_point_println(candidate, 1); );
+       DOLOG_SHORT( replace_sourcestate_println_short(ss); );
 
-                       replace_push_activation_record(es, candidate, prevframe, code, ss.frames);
-                       parent = NULL;
-               }
-               else {
-                       parent = candidate;
+#if !defined(NDEBUG)
+       /* avoid infinite loops by self-replacement, only if not in testing mode */
+
+       if (!opt_TestReplacement) {
+               frame = ss->frames;
+               while (frame->down)
+                       frame = frame->down;
+
+               if (frame->torp == origrp) {
+                       DOLOG_SHORT(
+                               printf("WARNING: identity replacement, turning off rps to avoid infinite loop\n");
+                       );
+                       replace_deactivate_replacement_points(origcode);
                }
-               DOLOG( replace_executionstate_println(es); );
        }
+#endif
 
-       DOLOG( replace_executionstate_println(es); );
-
-       assert(candidate);
-       if (candidate == origrp) {
-               DOLOG_SHORT(
-                       printf("WARNING: identity replacement, turning off rp to avoid infinite loop\n");
-               );
-               replace_deactivate_replacement_point(origrp);
+#if defined(ENABLE_THREADS) && defined(ENABLE_GC_CACAO)
        }
+#endif
+
+       /* build the new execution state */
+
+       replace_build_execution_state(ss, es);
+
+#if !defined(NDEBUG)
+       /* continue execution after patched machine code, if testing mode enabled */
+
+       if (opt_TestReplacement)
+               es->pc += REPLACEMENT_PATCH_SIZE;
+#endif
 
        /* release dump area */
 
        dump_release(dumpsize);
+}
+
+
+/* replace_me_wrapper **********************************************************
+
+   This function is called by the signal handler. It determines if there
+   is an active replacement point pending at the given PC and returns
+   accordingly.
+
+   THIS FUNCTION MUST BE CALLED USING A SAFE STACK AREA!
+
+   IN:
+       pc...............the program counter that triggered the replacement.
+       context..........the context (machine state) to which the
+                           replacement should be applied.
+
+   OUT:
+       context..........the context after replacement finished.
+
+   RETURN VALUE:
+       true.............replacement done, everything went ok
+       false............no replacement done, context unchanged
+
+*******************************************************************************/
+
+bool replace_me_wrapper(u1 *pc, void *context)
+{
+       codeinfo         *code;
+       rplpoint         *rp;
+       executionstate_t  es;
+
+       /* search the codeinfo for the given PC */
+
+       code = code_find_codeinfo_for_pc(pc);
+       assert(code);
+
+       /* search for a replacement point at the given PC */
+
+       rp = replace_find_replacement_point_for_pc(code, pc);
+
+       /* check if the replacement point belongs to given PC and is active */
+
+       if ((rp != NULL) && (rp->pc == pc) && (rp->flags & RPLPOINT_FLAG_ACTIVE)) {
+
+               /* set codeinfo pointer in execution state */
+
+               es.code = code;
+
+               /* read execution state from current context */
+
+               md_replace_executionstate_read(&es, context);
+
+               DOLOG( printf("REPLACEMENT READ: ");
+                          replace_executionstate_println(&es); );
+
+               /* do the actual replacement */
+
+               replace_me(rp, &es);
+
+               /* write execution state to current context */
+
+               md_replace_executionstate_write(&es, context);
+
+               DOLOG( printf("REPLACEMENT WRITE: ");
+                          replace_executionstate_println(&es); );
 
-       /* enter new code */
+               /* new code is entered after returning */
 
-       DOLOG( printf("JUMPING IN!\n"); fflush(stdout); );
+               DOLOG( printf("JUMPING IN!\n"); fflush(stdout); );
+               return true;
+       }
+       else
+               return false;
+}
+
+
+/******************************************************************************/
+/* NOTE: Stuff specific to the exact GC is below.                             */
+/******************************************************************************/
+
+#if defined(ENABLE_GC_CACAO)
+void replace_gc_from_native(threadobject *thread, u1 *pc, u1 *sp)
+{
+       stackframeinfo_t *sfi;
+       executionstate_t *es;
+       sourcestate_t    *ss;
+
+       /* get the stackframeinfo of this thread */
+       assert(thread == THREADOBJECT);
+       sfi = STACKFRAMEINFO;
+
+       /* create the execution state */
+       es = DNEW(executionstate_t);
+       es->pc = pc;
+       es->sp = sp;
+       es->pv = 0;      /* since we are in a native, PV is invalid! */
+       es->code = NULL; /* since we are in a native, we do not have a codeinfo */
+
+       /* we assume we are in a native (no replacement point)! */
+       ss = replace_recover_source_state(NULL, sfi, es);
+
+       /* map the sourcestate using the identity mapping */
+       replace_map_source_state_identity(ss);
+
+       /* remember executionstate and sourcestate for this thread */
+       GC_EXECUTIONSTATE = es;
+       GC_SOURCESTATE    = ss;
+}
+#endif
+
+#if defined(ENABLE_GC_CACAO)
+void replace_gc_into_native(threadobject *thread)
+{
+       executionstate_t *es;
+       sourcestate_t    *ss;
 
-#if (defined(__I386__) || defined(__X86_64__) || defined(__ALPHA__) || defined(__POWERPC__) || defined(__MIPS__)) && defined(ENABLE_JIT)
-       asm_replacement_in(es);
+       /* get the executionstate and sourcestate for the given thread */
+       es = GC_EXECUTIONSTATE;
+       ss = GC_SOURCESTATE;
+
+       /* rebuild the stack of the given thread */
+       replace_build_execution_state(ss, es);
+}
 #endif
-       abort(); /* NOT REACHED */
+
+
+/******************************************************************************/
+/* NOTE: No important code below.                                             */
+/******************************************************************************/
+
+
+/* statistics *****************************************************************/
+
+#if defined(REPLACE_STATISTICS)
+static void print_freq(FILE *file,int *array,int limit)
+{
+       int i;
+       int sum = 0;
+       int cum = 0;
+       for (i=0; i<limit; ++i)
+               sum += array[i];
+       sum += array[limit];
+       for (i=0; i<limit; ++i) {
+               cum += array[i];
+               fprintf(file,"      %3d: %8d (cum %3d%%)\n",
+                               i, array[i], (sum) ? ((100*cum)/sum) : 0);
+       }
+       fprintf(file,"    >=%3d: %8d\n",limit,array[limit]);
+}
+#endif /* defined(REPLACE_STATISTICS) */
+
+
+#if defined(REPLACE_STATISTICS)
+
+#define REPLACE_PRINT_DIST(name, array)                              \
+    printf("    " name " distribution:\n");                          \
+    print_freq(stdout, (array), sizeof(array)/sizeof(int) - 1);
+
+void replace_print_statistics(void)
+{
+       printf("replacement statistics:\n");
+       printf("    # of replacements:   %d\n", stat_replacements);
+       printf("    # of frames:         %d\n", stat_frames);
+       printf("    # of recompilations: %d\n", stat_recompile);
+       printf("    patched static calls:%d\n", stat_staticpatch);
+       printf("    unrolled inlines:    %d\n", stat_unroll_inline);
+       printf("    unrolled calls:      %d\n", stat_unroll_call);
+       REPLACE_PRINT_DIST("frame depth", stat_dist_frames);
+       REPLACE_PRINT_DIST("locals per frame", stat_dist_locals);
+       REPLACE_PRINT_DIST("ADR locals per frame", stat_dist_locals_adr);
+       REPLACE_PRINT_DIST("primitive locals per frame", stat_dist_locals_prim);
+       REPLACE_PRINT_DIST("RET locals per frame", stat_dist_locals_ret);
+       REPLACE_PRINT_DIST("void locals per frame", stat_dist_locals_void);
+       REPLACE_PRINT_DIST("stack slots per frame", stat_dist_stack);
+       REPLACE_PRINT_DIST("ADR stack slots per frame", stat_dist_stack_adr);
+       REPLACE_PRINT_DIST("primitive stack slots per frame", stat_dist_stack_prim);
+       REPLACE_PRINT_DIST("RET stack slots per frame", stat_dist_stack_ret);
+       printf("\n");
+       printf("    # of methods:            %d\n", stat_methods);
+       printf("    # of replacement points: %d\n", stat_rploints);
+       printf("    # of regallocs:          %d\n", stat_regallocs);
+       printf("        per rplpoint:        %f\n", (double)stat_regallocs / stat_rploints);
+       printf("        per method:          %f\n", (double)stat_regallocs / stat_methods);
+       REPLACE_PRINT_DIST("replacement points per method", stat_dist_method_rplpoints);
+       printf("\n");
+
+}
+#endif /* defined(REPLACE_STATISTICS) */
+
+
+#if defined(REPLACE_STATISTICS)
+static void replace_statistics_source_frame(sourceframe_t *frame)
+{
+       int adr = 0;
+       int ret = 0;
+       int prim = 0;
+       int vd = 0;
+       int n = 0;
+       int i;
+
+       for (i=0; i<frame->javalocalcount; ++i) {
+               switch (frame->javalocaltype[i]) {
+                       case TYPE_ADR: adr++; break;
+                       case TYPE_RET: ret++; break;
+                       case TYPE_INT: case TYPE_LNG: case TYPE_FLT: case TYPE_DBL: prim++; break;
+                       case TYPE_VOID: vd++; break;
+                       default: assert(0);
+               }
+               n++;
+       }
+       REPLACE_COUNT_DIST(stat_dist_locals, n);
+       REPLACE_COUNT_DIST(stat_dist_locals_adr, adr);
+       REPLACE_COUNT_DIST(stat_dist_locals_void, vd);
+       REPLACE_COUNT_DIST(stat_dist_locals_ret, ret);
+       REPLACE_COUNT_DIST(stat_dist_locals_prim, prim);
+       adr = ret = prim = n = 0;
+       for (i=0; i<frame->javastackdepth; ++i) {
+               switch (frame->javastacktype[i]) {
+                       case TYPE_ADR: adr++; break;
+                       case TYPE_RET: ret++; break;
+                       case TYPE_INT: case TYPE_LNG: case TYPE_FLT: case TYPE_DBL: prim++; break;
+               }
+               n++;
+       }
+       REPLACE_COUNT_DIST(stat_dist_stack, n);
+       REPLACE_COUNT_DIST(stat_dist_stack_adr, adr);
+       REPLACE_COUNT_DIST(stat_dist_stack_ret, ret);
+       REPLACE_COUNT_DIST(stat_dist_stack_prim, prim);
 }
+#endif /* defined(REPLACE_STATISTICS) */
 
 
+/* debugging helpers **********************************************************/
+
 /* replace_replacement_point_println *******************************************
  
    Print replacement point info.
@@ -1955,6 +3106,16 @@ void replace_me(rplpoint *rp, executionstate_t *es)
 
 #define TYPECHAR(t)  (((t) >= 0 && (t) <= TYPE_RET) ? show_jit_type_letters[t] : '?')
 
+static char *replace_type_str[] = {
+       "STD",
+       "EXH",
+       "SBR",
+       "CALL",
+       "INLINE",
+       "RETURN",
+       "BODY"
+};
+
 void replace_replacement_point_println(rplpoint *rp, int depth)
 {
        int j;
@@ -1968,11 +3129,15 @@ void replace_replacement_point_println(rplpoint *rp, int depth)
        for (j=0; j<depth; ++j)
                putchar('\t');
 
-       printf("rplpoint (id %d) %p pc:%p+%d out:%p target:%p mcode:%016llx type:%s",
-                       rp->id, (void*)rp,rp->pc,rp->callsize,rp->outcode,(void*)rp->target,
-                       (unsigned long long)rp->mcode,replace_type_str[rp->type]);
+       printf("rplpoint (id %d) %p pc:%p+%d type:%s",
+                       rp->id, (void*)rp,rp->pc,rp->callsize,
+                       replace_type_str[rp->type]);
        if (rp->flags & RPLPOINT_FLAG_NOTRAP)
                printf(" NOTRAP");
+       if (rp->flags & RPLPOINT_FLAG_COUNTDOWN)
+               printf(" COUNTDOWN");
+       if (rp->flags & RPLPOINT_FLAG_ACTIVE)
+               printf(" ACTIVE");
        printf(" parent:%p\n", (void*)rp->parent);
        for (j=0; j<depth; ++j)
                putchar('\t');
@@ -2005,7 +3170,7 @@ void replace_replacement_point_println(rplpoint *rp, int depth)
 
        printf("\n");
 }
-#endif
+#endif /* !defined(NDEBUG) */
 
 
 /* replace_show_replacement_points *********************************************
@@ -2035,6 +3200,9 @@ void replace_show_replacement_points(codeinfo *code)
        printf("\ttotal allocations : %d\n",code->regalloccount);
        printf("\tsaved int regs    : %d\n",code->savedintcount);
        printf("\tsaved flt regs    : %d\n",code->savedfltcount);
+#if defined(HAS_ADDRESS_REGISTER_FILE)
+       printf("\tsaved adr regs    : %d\n",code->savedadrcount);
+#endif
        printf("\tmemuse            : %d\n",code->memuse);
 
        printf("\n");
@@ -2042,8 +3210,6 @@ void replace_show_replacement_points(codeinfo *code)
        for (i=0; i<code->rplpointcount; ++i) {
                rp = code->rplpoints + i;
 
-               assert(rp->code == code);
-
                depth = 1;
                parent = rp->parent;
                while (parent) {
@@ -2072,7 +3238,7 @@ void replace_executionstate_println(executionstate_t *es)
        int slots;
        stackslot_t *sp;
        int extraslots;
-       
+
        if (!es) {
                printf("(executionstate_t *)NULL\n");
                return;
@@ -2088,7 +3254,11 @@ void replace_executionstate_println(executionstate_t *es)
                        printf("\t");
                else
                        printf(" ");
-               printf("%-3s = %016llx",regs[i],(unsigned long long)es->intregs[i]);
+#if SIZEOF_VOID_P == 8
+               printf("%-3s = %016llx",abi_registers_integer_name[i],(unsigned long long)es->intregs[i]);
+#else
+               printf("%-3s = %08lx",abi_registers_integer_name[i],(unsigned long)es->intregs[i]);
+#endif
                if (i%4 == 3)
                        printf("\n");
        }
@@ -2101,6 +3271,17 @@ void replace_executionstate_println(executionstate_t *es)
                if (i%4 == 3)
                        printf("\n");
        }
+# if defined(HAS_ADDRESS_REGISTER_FILE)
+       for (i=0; i<ADR_REG_CNT; ++i) {
+               if (i%4 == 0)
+                       printf("\t");
+               else
+                       printf(" ");
+               printf("A%02d = %016llx",i,(unsigned long long)es->adrregs[i]);
+               if (i%4 == 3)
+                       printf("\n");
+       }
+# endif
 #endif
 
        sp = (stackslot_t *) es->sp;
@@ -2109,7 +3290,7 @@ void replace_executionstate_println(executionstate_t *es)
 
        if (es->code) {
                methoddesc *md = es->code->m->parseddesc;
-               slots = code_get_stack_frame_size(es->code);
+               slots = es->code->stackframesize;
                extraslots = 1 + md->memuse;
        }
        else
@@ -2144,32 +3325,41 @@ void replace_executionstate_println(executionstate_t *es)
 #endif
 
 #if !defined(NDEBUG)
-void java_value_print(s4 type, u8 value)
+static void java_value_print(s4 type, replace_val_t value)
 {
-       java_objectheader *obj;
-       utf               *u;
+       java_object_t *obj;
+       utf           *u;
 
-       printf("%016llx",(unsigned long long) value);
+       printf("%016llx",(unsigned long long) value.l);
 
        if (type < 0 || type > TYPE_RET)
                printf(" <INVALID TYPE:%d>", type);
        else
                printf(" %s", show_jit_type_names[type]);
 
-       if (type == TYPE_ADR && value != 0) {
-               obj = (java_objectheader *) (ptrint) value;
+       if (type == TYPE_ADR && value.a != NULL) {
+               obj = value.a;
                putchar(' ');
                utf_display_printable_ascii_classname(obj->vftbl->class->name);
 
                if (obj->vftbl->class == class_java_lang_String) {
                        printf(" \"");
-                       u = javastring_toutf((java_lang_String *)obj, false);
+                       u = javastring_toutf(obj, false);
                        utf_display_printable_ascii(u);
                        printf("\"");
                }
        }
-       else if (type == TYPE_INT || type == TYPE_LNG) {
-               printf(" %lld", (long long) value);
+       else if (type == TYPE_INT) {
+               printf(" %ld", (long) value.i);
+       }
+       else if (type == TYPE_LNG) {
+               printf(" %lld", (long long) value.l);
+       }
+       else if (type == TYPE_FLT) {
+               printf(" %f", value.f);
+       }
+       else if (type == TYPE_DBL) {
+               printf(" %f", value.d);
        }
 }
 #endif /* !defined(NDEBUG) */
@@ -2178,16 +3368,38 @@ void java_value_print(s4 type, u8 value)
 #if !defined(NDEBUG)
 void replace_source_frame_println(sourceframe_t *frame)
 {
-       s4 i;
+       s4 i,j;
        s4 t;
 
+       if (REPLACE_IS_NATIVE_FRAME(frame)) {
+               printf("\tNATIVE\n");
+               printf("\tsfi: "); replace_stackframeinfo_println(frame->sfi);
+               printf("\tnativepc: %p\n", frame->nativepc);
+               printf("\tframesize: %d\n", frame->nativeframesize);
+
+               j = 0;
+               for (i=0; i<INT_REG_CNT; ++i) {
+                       if (nregdescint[i] == REG_SAV)
+                               printf("\t%s = %p\n", abi_registers_integer_name[i], (void*)frame->nativesavint[j++]);
+               }
+
+               j = 0;
+               for (i=0; i<FLT_REG_CNT; ++i) {
+                       if (nregdescfloat[i] == REG_SAV)
+                               printf("\tF%02d = %f\n", i, frame->nativesavflt[j++]);
+               }
+
+               printf("\n");
+               return;
+       }
+
        printf("\t");
        method_println(frame->method);
        printf("\tid: %d\n", frame->id);
        printf("\ttype: %s\n", replace_type_str[frame->type]);
        printf("\n");
 
-       if (frame->instance) {
+       if (frame->instance.a) {
                printf("\tinstance: ");
                java_value_print(TYPE_ADR, frame->instance);
                printf("\n");
@@ -2230,19 +3442,33 @@ void replace_source_frame_println(sourceframe_t *frame)
                for (i=0; i<frame->syncslotcount; ++i) {
                        printf("\tslot[%2d] = ",i);
 #ifdef HAS_4BYTE_STACKSLOT
-                       printf("%08lx\n",(unsigned long) frame->syncslots[i]);
+                       printf("%08lx\n",(unsigned long) frame->syncslots[i].p);
 #else
-                       printf("%016llx\n",(unsigned long long) frame->syncslots[i]);
+                       printf("%016llx\n",(unsigned long long) frame->syncslots[i].p);
 #endif
                }
                printf("\n");
        }
 
-       if (frame->readrp) {
-               printf("\tread replacement point:\n");
-               replace_replacement_point_println(frame->readrp, 2);
-               printf("\n");
+       if (frame->fromcode) {
+               printf("\tfrom %p ", (void*)frame->fromcode);
+               method_println(frame->fromcode->m);
+       }
+       if (frame->tocode) {
+               printf("\tto %p ", (void*)frame->tocode);
+               method_println(frame->tocode->m);
+       }
+
+       if (frame->fromrp) {
+               printf("\tfrom replacement point:\n");
+               replace_replacement_point_println(frame->fromrp, 2);
        }
+       if (frame->torp) {
+               printf("\tto replacement point:\n");
+               replace_replacement_point_println(frame->torp, 2);
+       }
+
+       printf("\n");
 }
 #endif /* !defined(NDEBUG) */
 
@@ -2276,6 +3502,66 @@ void replace_sourcestate_println(sourcestate_t *ss)
 }
 #endif
 
+
+/* replace_sourcestate_println_short *******************************************
+
+   Print a compact representation of the given source state.
+
+   IN:
+       ss...............the source state to print
+
+*******************************************************************************/
+
+#if !defined(NDEBUG)
+void replace_sourcestate_println_short(sourcestate_t *ss)
+{
+       sourceframe_t *frame;
+
+       for (frame = ss->frames; frame != NULL; frame = frame->down) {
+               printf("\t");
+
+               if (REPLACE_IS_NATIVE_FRAME(frame)) {
+                       printf("NATIVE (pc %p size %d) ",
+                                       (void*)frame->nativepc, frame->nativeframesize);
+                       replace_stackframeinfo_println(frame->sfi);
+                       continue;
+               }
+
+               if (frame->torp) {
+                       printf("%c", (frame->torp == frame->fromrp) ? '=' : '+');
+               }
+
+               printf("%s", replace_type_str[frame->fromrp->type]);
+
+               if (frame->torp && frame->torp->type != frame->fromrp->type)
+                       printf("->%s", replace_type_str[frame->torp->type]);
+
+               if (frame->tocode != frame->fromcode)
+                       printf(" (%p->%p/%d) ",
+                                  (void*) frame->fromcode, (void*) frame->tocode,
+                                  frame->fromrp->id);
+               else
+                       printf(" (%p/%d) ", (void*) frame->fromcode, frame->fromrp->id);
+
+               method_println(frame->method);
+       }
+}
+#endif
+
+#if !defined(NDEBUG)
+static void replace_stackframeinfo_println(stackframeinfo_t *sfi)
+{
+       printf("prev=%p pv=%p sp=%p ra=%p xpc=%p method=",
+                       (void*)sfi->prev, (void*)sfi->pv, (void*)sfi->sp,
+                       (void*)sfi->ra, (void*)sfi->xpc);
+
+       if (sfi->code)
+               method_println(sfi->code->m);
+       else
+               printf("(nil)\n");
+}
+#endif
+
 /*
  * These are local overrides for various environment variables in Emacs.
  * Please do not remove this and leave it at the end of the file, where